In 24 months, Cursor went from nothing to $2 billion in annual recurring revenue. For context, Slack took five years to hit $1 billion. Zoom took nine. Cursor doubled past that number in a single quarter - from $1 billion in November 2025 to $2 billion by February 2026.

And yet, a familiar question follows the company: is Cursor anything more than a really good wrapper?

The “wrapper” critique is worth taking seriously. Cursor’s intelligence layer runs on models from OpenAI, Anthropic, and Google - companies that are also building competing coding tools. Its editor is a fork of VS Code, Microsoft’s free, open-source code editor that already dominates 70% of the developer market. The skeptic’s version: strip away the borrowed brains and the borrowed interface, and what’s left?

As of March 12, 2026, Cursor is in talks for a $50 billion valuation. That’s 25 times revenue for a company whose core dependencies are controlled by its competitors. The bulls see a company redefining how software gets built. The bears see a house built on rented land. Both have evidence on their side.

I think the “wrapper” frame misses something fundamental. The real question isn’t whether Cursor depends on external models - it clearly does, and that’s a feature as much as a risk. It’s whether the integration layer they’ve built on top of that intelligence is valuable enough to stand on its own. That tension, what I’ll call the integration tax, is the defining challenge of every AI startup building on foundation models.

TLDR

• Cursor reportedly hit $2B ARR in ~24 months, the fastest revenue ramp in SaaS history, doubling from $1B to $2B in just three months

• The VS Code fork was the defining decision. By forking the open-source editor instead of building an extension, Cursor inherited the world’s largest developer ecosystem while gaining total control over the AI experience

• “Wrapper” is the wrong frame. Cursor is building a proprietary model (Composer), orchestrating multiple AI providers, and locking in enterprises through integrations. But its moat is thinner than bulls want to believe

• Enterprise revenue is now ~60% of total, with 64% of Fortune 500 companies using Cursor and deployments at Stripe (80% adoption), Infosys (100,000 engineers), and Nvidia (30,000 engineers)

• The competitive threats are real but navigable. Claude Code, OpenAI Codex, and autonomous agents challenge the IDE form factor, and Microsoft can restrict fork access - but Cursor has already adapted to marketplace lockouts and continues to outpace competitors on product velocity

• The “integration tax” is the defining question of the AI era. Every startup building on top of foundation models will eventually face Cursor’s dilemma: is the orchestration layer worth more than the intelligence it orchestrates?

Four Kids and a Fork

Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger met at MIT. They were math olympiad competitors, hackathon addicts, and, by their own admission, initially terrible at picking what to build.

Their first idea was a 3D CAD tool. Truell later described it bluntly: “It was a bad idea. The founder-market fit was horrible.”

What they did understand was developer tools. And by 2022, they saw a gap that the incumbents were ignoring. AI was getting good enough to transform how code gets written, but no one was building a code editor designed for AI from the ground up. GitHub Copilot existed, but it was bolted onto VS Code as an extension, sandboxed and limited in what it could see and do. Like putting a jet engine on a bicycle.

So they made the decision that would define everything that followed. Instead of building a VS Code extension (the safe, conventional path) or creating an entirely new editor (the ambitious, slow path), they forked VS Code.

This matters more than it sounds. VS Code is MIT-licensed open source. Anyone can take the code and build on it. But what Cursor’s fork gave them was something no extension could achieve: total control over the editing experience. They could rewrite the rendering pipeline, embed AI into every interaction, build features that were architecturally impossible from inside an extension sandbox.

“We were really, really intentional about wanting to own the surface,” Truell said in a Founder Boat interview.

And because they forked VS Code rather than starting from scratch, every developer could import their settings, extensions, themes, and keybindings in minutes. Near-zero switching cost with a radically better product underneath.

The fork decision was Cursor’s version of renting the ecosystem while owning the experience. It worked.

The Growth That Broke Every Record

The numbers are hard to overstate.

No SaaS company has ever done this. A caveat: SaaS adoption velocity has structurally accelerated since the 2010s. PLG is more mature, cloud-native is the default, and developer communities spread tools faster than they did when Slack launched in 2014. Even so, the gap is wild. Slack needed three years to reach $100M ARR. Cursor blew past that in twelve months.

The valuation has followed:

• $400 million in August 2024

• $2.6 billion in December 2024

• $9.9 billion in June 2025

• $29.3 billion in November 2025.

A 67x increase in sixteen months. And now $50 billion is on the table.

All of this with roughly 150 to 182 employees. That works out to somewhere between $5.5 million and $11 million in revenue per employee - numbers that would make most enterprise software companies weep.

But raw growth tells you what happened. It doesn’t tell you whether it’s defensible.

Subscribe now

Utilising the Ecosystem, Owning the Experience

The fork decision wasn’t just clever. It was the foundation of every advantage Cursor has built since.

The mechanics matter. Cursor takes the VS Code codebase and layers AI throughout the entire editing experience. Not as a sidebar chatbot. Not as an autocomplete suggestion. AI is embedded into tab completions, inline edits, multi-file refactoring, codebase-wide search, and an agent mode that can spin up parallel coding agents working across your project simultaneously.

Cursor indexes your entire repository locally. This is what separates it from every chatbot wrapper on the market. The AI gets what engineers call “codebase-wide understanding.” It doesn’t just see the file you have open. It understands how that file connects to everything else. This is something raw model APIs simply cannot do. It’s the difference between a doctor reading one page of your chart and a doctor who has reviewed your entire medical history.

But there’s a cost to the fork strategy. And Microsoft is making Cursor pay it.

Here’s a useful analogy. Think of VS Code’s extension marketplace as an app store. Cursor forked the phone, but Microsoft still owns the store. And in April 2025, Microsoft started checking IDs at the door.

Here’s what happened: VS Code extensions are like apps that add features to your editor - language support, debugging tools, formatting. Microsoft runs the main store where developers download these extensions. Cursor had been quietly accessing that store through a workaround, even though Microsoft’s terms technically restricted it to their own products. In April 2025, Microsoft started enforcing those terms, blocking certain extensions from running in VS Code forks like Cursor.

Cursor adapted quickly. They moved to OpenVSX, an open-source alternative registry (think of it as a third-party app store), and started publishing their own versions of popular extensions. But a deeper asymmetry remains. Microsoft controls both VS Code and GitHub Copilot, which means Copilot gets access to premium, behind-the-scenes features that Cursor simply can’t use.

This is the fork tax. You inherit the ecosystem, but you don’t control the landlord. The good news: Cursor has shown it can move fast when the rules change. The risk: the landlord is also building a competing product.

The $50 Billion Wrapper Question

This is where the Cursor story splits in two. The company is simultaneously the best argument for and against building on top of foundation models.

More Than a Wrapper

Building Its Own Brain

Cursor isn’t just piping API calls to OpenAI and Claude and calling it a day. They’ve built Composer, a proprietary mixture-of-experts model trained through reinforcement learning inside real codebases. The model learned to use developer tools - semantic search, file editors, terminal commands - the way a developer actually works.

They’ve also built a multi-model orchestration layer. Users can choose between GPT-5.4, Claude Opus 4.6, Gemini 3 Pro, Grok Code, and Cursor’s own models. Cursor isn’t dependent on any single provider. It’s the switchboard.

The Workflow Moat

And then there’s the integration moat. Cursor now connects to Slack, Figma, and Linear. Within one week of the Linear integration launching, 45% of enterprise customers were assigning tasks to Cursor agents directly from their project management tool. That’s not a wrapper. That’s a workflow.

The “Bigger IDE” Thesis

Andrej Karpathy, the former Tesla AI chief who coined the term “vibe coding” while using Cursor’s Composer, posted a thread in March 2026 that reframed the entire debate. His argument: the age of the IDE isn’t over. We’re going to need a bigger IDE.

Karpathy went further, describing the IDE of the future as an “agent command center” - a place where you manage teams of AI agents, monitor their status, and orchestrate their work. He argued that agent structures will become “org code” - versionable, forkable, shareable. “You can’t fork classical orgs (eg Microsoft) but you’ll be able to fork agentic orgs.”

If Karpathy is right, Cursor isn’t a wrapper. It’s the operating system for a new kind of software development. And it’s further along that path than anyone else.

The Supplier Problem

Now, the tension at the heart of the business.

Cursor’s intelligence layer depends on models from OpenAI, Anthropic, and Google - companies that are also investing heavily in the coding tool layer. OpenAI tried to buy Windsurf for $3 billion in 2025 (the deal collapsed in July). Anthropic ships Claude Code, a terminal-based coding agent. Google paid $2.4 billion for the Windsurf team - and also invested in Cursor’s Series D. Everyone is hedging, which cuts both ways: these companies are both competitors and investors in Cursor’s success.

The bigger structural question is whether agents will eventually outgrow the IDE. If AI becomes capable enough to write code autonomously - reading specs, running tests, deploying without a human in the loop - then the text editor becomes less central. Claude Code and OpenAI’s Codex are exploring this future. Codex already has 1.6 million weekly active users. But “the IDE dies” has been predicted before, and developers keep gravitating back to rich editing environments. Karpathy’s “bigger IDE” thesis suggests the form factor evolves rather than disappears.

Margins, Churn, and the Pricing Tightrope

The Margin Unknown

Cursor doesn’t disclose gross margins, and this is the question that will ultimately separate a $50 billion company from a $15 billion one. AI API costs plus custom model training infrastructure are expensive. Traditional SaaS companies operate at 70%+ gross margins. AI-heavy products may compress to 50-60%.

Run the numbers at $2 billion in revenue. At a traditional SaaS margin of 75%, that’s $1.5 billion in gross profit. At a compressed AI margin of 55%, it’s $1.1 billion. That $400 million gap matters because at 25x revenue, the market is pricing Cursor as though it will eventually reach SaaS-like margins. There’s a plausible path: Composer reduces reliance on expensive third-party API calls, and scale drives down per-user costs. But that path hasn’t been proven yet.

Pricing Growing Pains

The August 2025 pricing backlash is worth understanding in context. Cursor switched to a credit-based billing system that effectively cut the number of AI requests users could make. Users revolted. The CEO had to publicly apologize. Some migrated to Windsurf. Cursor’s own AI support bot made things worse by hallucinating a false policy during the crisis. It was messy - but the company course-corrected, and growth actually accelerated afterward. The incident revealed a product people cared enough to fight over, not one they quietly abandoned.

The Churn Paradox

Churn runs hot: industry estimates put blended monthly churn at 3-5%, which annualizes to 30-46%. For context, high individual churn is typical for PLG tools in hypergrowth - Slack and Dropbox saw similar patterns before enterprise revenue stabilized their base. At $2 billion ARR, that means replacing $600 million to $920 million in revenue every year just to stay flat.

But there’s an important nuance. If enterprise accounts (60% of revenue) are as sticky as the deployment data suggests, then individual developer churn must be significantly worse than the blended number. Pro users at $20/month are the ones churning at high rates, while enterprise contracts hold. That’s actually a more manageable problem than it first appears. Enterprise is where the money concentrates, and Cursor’s enterprise mix is growing. The consumer tier increasingly serves as a free trial funnel rather than a revenue center in its own right.

The Enterprise Blitz

If Cursor has a moat, enterprise is where it’s being poured. And the way it got there tells you something about how durable it might be.

According to industry estimates, about 60% of Cursor’s revenue now comes from enterprise customers. (Cursor hasn’t confirmed this figure publicly.) That shift happened in roughly twelve months. But the interesting part isn’t the speed. It’s the mechanism.

Cursor didn’t sell its way into these accounts. It grew from the bottom up. A developer tries the free tier (2,000 AI completions). Gets hooked. Their teammates notice. Word spreads. Then IT gets a call from a hundred engineers all asking for the same tool. By the time procurement is involved, the decision is already made.

Stripe’s adoption followed exactly this arc. It went from single digits to over 80% of engineers using Cursor, all through organic adoption. No enterprise sales team knocking on doors. That pattern matters because it means Cursor isn’t buying its enterprise base with expensive sales motions. The product is doing the selling.

The question is whether that bottom-up adoption translates into top-down lock-in. Infosys’s answer suggests it can. They deployed Cursor across a Center of Excellence serving 100,000 engineers and reported a 39% increase in shipped pull requests. (More PRs shipped doesn’t automatically mean more productivity. PRs could be smaller or lower quality. But the directional signal is hard to ignore at that scale.) When a tool measurably increases output at that scale, the switching cost becomes organizational, not just technical. Nvidia’s 30,000-engineer deployment reportedly tripled code output. At that point, removing Cursor means explaining to the board why engineering velocity just dropped.

Industry analysts estimate that 64% of Fortune 500 companies now use Cursor in some capacity, though this figure hasn’t been independently verified. SOC 2 compliance, audit logs, centralized billing. These are the features that turn a popular tool into an embedded one.

The competitive picture matters. GitHub Copilot still holds 42% of the paid AI coding tool market versus Cursor’s 18%. And Copilot is no longer just autocomplete.

Copilot Workspace and Agent Mode now offer multi-file editing, codebase-aware suggestions, and autonomous task execution inside native VS Code. These features directly compete with Cursor’s core differentiators, not just on price but on capability.

Microsoft could also bundle Copilot free with VS Code and GitHub - though they haven’t pulled that trigger yet, possibly because paid Copilot is too profitable to cannibalize. Cursor’s edge has been speed of innovation: every major Copilot feature update has trailed a Cursor release by months. The question is whether that lead is structural or just a function of startup velocity.

What Would Have to Be True

For the $50 billion valuation to hold, several things need to work at once.

• Growth sustaining above 50% annually for three more years, pushing ARR toward $8 billion. Enterprise retention exceeding 90% net revenue retention.

• Composer meaningfully reducing dependence on external providers.

• The integrations creating real switching costs.

• And most fundamentally, the IDE remaining the central interface for software development, even as agents try to replace it.

None of these are implausible. Cursor has momentum on every front.

The bear case is that any single disruption could compress the trajectory:

• Microsoft bundles Copilot free

• Autonomous agents reduce the editor’s role

• A model provider ships a good-enough IDE

But Cursor has already navigated several of these threats (the marketplace lockout, the pricing backlash, the Windsurf competition) and come out stronger each time. The question isn’t whether challenges will come. It’s whether the team that’s outexecuted everyone so far can keep doing it.

Cursor has built something genuinely impressive - and the market is pricing in a future that still has to be earned.

The Integration Tax

Every AI startup will eventually face Cursor’s dilemma. If you build on top of foundation models, your suppliers are also potential competitors. Your product’s intelligence is rented - but that doesn’t make the product any less valuable to the people using it.

The core question of the integration layer adds enough value to stand on its own as intelligence commoditizes lingers. The UX polish, the enterprise lock-in, the data flywheel from a million daily users, the proprietary models like Composer. Does all of that compound into something defensible?

I call this the integration tax. Strategists will recognize echoes of Ben Thompson’s Aggregation Theory and Christensen’s value chain disruption here. But the AI version has a twist: in the AI stack, your commoditized supplier isn’t just a passive vendor. They’re actively building competing products. The tax comes due when your supplier decides to compete directly - and the bill is lower for companies that have built genuine integration depth.

We’ve seen this play out before. Zynga built a gaming empire on Facebook’s platform, reaching 300 million monthly active users by 2012. When Facebook changed its algorithm and throttled viral distribution, Zynga lost 85% of its market cap in 18 months. The integration was real. The switching costs were real. But the platform controlled the oxygen supply.

The test for any AI startup is simple: if the model provider launched a competing product tomorrow, would your customers stay? If the answer depends on workflow integrations, enterprise compliance, data advantages, you likely have a durable business. If the answer depends entirely on your product being “better right now,” the position is fragile.

Cursor passes more of this test than its critics acknowledge. The fork gives them UX control. Composer gives them growing intelligence independence. Enterprise gives them genuine switching costs. The foundation uses rented intelligence, yes - but the building on top of it is increasingly Cursor’s own. And the landlords, for all their ambitions in coding tools, haven’t yet matched what Cursor delivers.

Any time you see an AI startup raising at 25x revenue while its core intelligence comes from a third party, run three checks.

1. First, what percentage of the product experience breaks if you swap in a different model? If the answer is “almost nothing,” the integration layer is doing real work. If the answer is “everything,” you’re looking at a wrapper.

2. Second, would the company’s enterprise customers need to retrain workflows, reconfigure compliance systems, or rebuild integrations to switch? The higher the organizational switching cost, the more durable the moat.

3. Third, does the company own any proprietary models or data assets that improve with scale? A pure orchestration layer without a learning loop is always at risk.

Cursor passes the second test and partially passes the first and third. That’s better than most AI startups, and the trajectory is improving - Composer and enterprise lock-in are actively strengthening those scores.

Cursor’s timer hasn’t run out. Whether it’s a countdown or a launchpad depends on what the company builds in the next twelve months. The founders who were smart enough to fork VS Code in 2022 might be smart enough to reinvent the IDE again. That’s the bet.

In October 2012, a man named Felix Baumgartner stood on a platform at the edge of space - nearly 39 kilometers above the New Mexico desert - and jumped. Eight million people watched live on YouTube. Every major news network covered it. The footage ran on a loop for weeks. And here’s the thing that still blows my mind: not a single second of it was a Red Bull commercial. It was all a Red Bull commercial.

The Stratos jump cost is estimated to be somewhere between $30 and $50 million. It generated an estimated $6 billion in earned media coverage - though “earned media value” is a notoriously unreliable metric, and a skeptical analyst would point out that the more verifiable outcome was a 7% bump in US sales over the six months that followed.

But the real return wasn’t in the quarter. It was in what the stunt revealed about a company that operates unlike anything else in consumer goods.

The Origin Story

In 1984, Dietrich Mateschitz was a 40-year-old toothpaste salesman for a German company called Blendax. On a business trip to Bangkok, he tried a syrupy Thai energy tonic called Krating Daeng - popular with truck drivers and factory workers - and noticed it cured his jet lag. He tracked down the manufacturer, Chaleo Yoovidhya, and the two each put up $500,000 to create Red Bull GmbH. Mateschitz adapted the formula for Western tastes - carbonated it, dialed back the sweetness - and prepared to launch in Austria.

The focus groups were a disaster. Consumers called the taste “disgusting.” The color was “unappetizing.” The research firm declared a total failure. Mateschitz read the same data and reached the opposite conclusion. Half the room loved it. Half hated it. He launched anyway, into a category that did not yet exist. The Energy Drink category.

From the start, Mateschitz made a structural choice that still defines the company. Red Bull doesn’t make Red Bull. Production has been outsourced to Rauch, an Austrian bottler, since 1987. The formula isn’t even proprietary. The liquid is a commodity. What Mateschitz kept in-house was everything else - the sales force, the marketing, the media, the events, the brand. He built a company where the product costs nine cents to manufacture and the entire business model sits on top of the markup.

And then he made the choice that would matter more than any other: he never took the company public. No shareholders. No quarterly earnings calls. No external capital. Just two families - Mateschitz and Yoovidhya - and the freedom to run the business on a 35-year time horizon.

The Twin Engine

Red Bull runs on two reinforcing engines: a marketing machine that builds appreciating assets, and a pricing structure that funds the whole thing. Neither engine works without the other.

Engine 1: Marketing That Compounds

Most consumer brands spend on impressions. A Super Bowl ad costs $7 million for 30 seconds. A billboard campaign runs for a month. A sponsored Instagram post lasts a scroll. These are depreciating expenditures - the moment the ad stops running, the value is gone.

Red Bull spends €3bn annually on assets. It owns:

• 2 F1 teams

• 4 primary football clubs across three continents - including RB Leipzig in the Bundesliga

• and sponsors 800-plus athletes.

Then there’s Red Bull Media House. Founded in 2007, it’s a full production company cranking out films, documentaries, and live event broadcasts for Red Bull TV, The Red Bulletin magazine, and streaming platforms. Redbull’s YouTube channel alone has over 27 million subscribers.

The difference matters because assets appreciate over time while impressions depreciate instantly. An F1 team generates exposure year after year. An event franchise like Red Bull Rampage or Red Bull Flugtag builds audience over decades. A media library grows. Meanwhile, that Super Bowl ad from three years ago? Nobody remembers it.

Mateschitz understood this explicitly. He once said, “In literal financial terms, our sports teams are not yet profitable, but in value terms, they are. The total editorial media value plus the media assets created around the teams are superior to pure advertising expenditures.”

Red Bull Racing posted a profit of just GBP 1.7 million in 2024 on an investment of $300-500 million, ROI of <1%. On paper, that 2024 return looks terrible. In practice, the F1 presence reaches over a billion viewers across 180-plus countries while simultaneously producing thousands of hours of content that Red Bull owns outright.

Red Bull doesn’t just sponsor events. It creates them. The Stratos jump. Flugtag. Crashed Ice. Cliff Diving. By owning the events, Red Bull retains full media rights. When every news outlet in the world covers the Stratos jump, they’re distributing Red Bull content for free. Red Bull turned media coverage into a multiplier, not a cost.

Engine 2: Premium Pricing

The asset-building machine only works because of a jaw-dropping margin structure. Red Bull’s production cost is estimated at roughly $0.09 per can. It wholesales for about $1.87 and retails for around $3.59 in Western markets. That’s a 40x markup from production to retail.

Red Bull charges about $0.40 per ounce. Monster charges about $0.20. Celsius charges about $0.25. Red Bull is roughly double the price per ounce of its closest competitor - for a product with nearly identical caffeine concentration (Red Bull: ~9.5mg per ounce, Monster: ~10mg per ounce).

Same caffeine. Double the price. The premium is 100% brand.

Mateschitz used premium pricing as a signal. A higher price communicates quality even when the product itself is chemically indistinguishable from cheaper alternatives. In blind taste tests, consumers often can’t tell Red Bull from generic energy drinks - but they rate the branded version as more energizing. Researchers have a name for this: placebo effect. The label changes the experience.

The small can reinforces the premium positioning. Red Bull’s standard is 8.4 fluid ounces. Monster’s standard is 16 ounces. Celsius is 12 ounces. Red Bull’s can is closer to a functional “dose” than a casual drink. You don’t nurse a Red Bull. You take it when you need energy now. The smaller format creates scarcity psychology and a “buy another” dynamic that larger cans don’t.

And Red Bull never discounts. When it raised distributor prices by 7% in September 2022, Monster’s stock price jumped 3% because analysts assumed Monster would follow suit. Red Bull’s pricing power is so strong that its decisions move competitors’ stocks.

Here’s why these two engines matter as a pair: That margin structure is what funds the $3 billion marketing machine. You can’t spend 25-30% of revenue on marketing if your margins are thin. Red Bull’s margins are enormous precisely because the brand premium is so high - and the brand premium is high precisely because of the marketing spend. Each reinforces the other. But this cycle doesn’t sustain itself automatically. It requires something most companies can’t replicate.

Subscribe now

The Private Advantage

Everything above, the F1 teams with dismal ROI, the five-year planning horizon for stunts like Stratos, the refusal to discount, the patience to reinvest all profits for the first 15 years of the company’s existence, is only possible because of one structural choice Mateschitz made in 1984 and never reversed: he kept Red Bull private.

Red Bull has never gone public. It has never taken outside capital. It has never issued debt. The company is split between two families - 49% held by Mark Mateschitz (inherited from his father Dietrich, who died in October 2022) and 51% held by the Yoovidhya family of Thailand.

Mateschitz was blunt about this. When asked about an IPO, he said: “It’s not a question of money. It’s a question of fun. Can you imagine me in a shareholders’ meeting?”

Another time: “A sale or an IPO is not even remotely tempting.”

And on the financial discipline that private ownership enables: “We at Red Bull spend the money that we’ve earned, not that we might earn someday.”

These quotes describe the philosophy the company was built.

Look at the timeline above. The Stratos jump took five years of development. The F1 program has been running for two decades. Red Bull Media House was built over 15 years into a full production company. None of these investments make sense on a quarterly reporting cycle. They only make sense when you’re measuring in decades.

Look at what happens to Red Bull’s public competitors.

Monster Beverage answers to Wall Street and to Coca-Cola, which holds a 21% stake. When Spruce Point Capital Management published a report in 2025 claiming Monster’s stock was overvalued, citing “diminishing US sales, slow international adoption, and increased competition,” Monster had to publicly defend its business. Red Bull’s growth can slow and speed up and slow again without anyone outside the two families ever knowing.

Celsius is even more instructive. After PepsiCo invested $550 million for an 8.5% stake in 2022, Celsius’s convenience store coverage exploded. Market share surged from under 1% to 10-16%. But by late 2024, Celsius stock was trading 70% below its May 2024 peak - not because the business was failing, but because growth was decelerating. The public market punished slowing growth even when the company was still growing.

A private Red Bull never faces this. It can invest in decade-long projects without the stock price acting as a live performance review.

Not Just an Energy Drink Story

Red Bull isn’t the only company that’s figured this out. The most ambitious companies of the last two decades have made the same structural choice - and for the same reason.

Stripe, the payments company valued at $159 billion as of February 2026, has stayed private for 16 years. Patrick Collison, its CEO, put it bluntly: “For us right now, an IPO would be a solution in search of a problem.” Stripe generated $2.2 billion in free cash flow in 2024 and processed $1.9 trillion in total payment volume in 2025. It doesn’t need public capital. What it needs is the freedom to build products on multi-year timelines without quarterly earnings pressure reshaping priorities - the same freedom Mateschitz needed to spend five years developing a space jump.

Databricks, now valued at $134 billion, watched its competitors rush to IPO during the 2021 boom - and then crash into cost-cutting mode when the market turned in 2022. While public companies were laying off engineers, Databricks hired thousands. CEO Ali Ghodsi was blunt about the timing: “It’s dumb to IPO this year,” he said in 2024, citing market volatility that would have forced short-term concessions on a company building for a decade-long AI infrastructure cycle.

SpaceX stayed private for over 23 years specifically so that quarterly earnings pressure couldn’t interfere with a Mars mission that might not generate returns for decades. Musk repeatedly insisted SpaceX would remain private as long as possible to protect its long-term ambitions, and wouldn’t consider going public “until you have an extremely stable and predictable revenue stream.” Only now - after Starlink revenue hit $10.4 billion and turned cash-flow positive - is a potential IPO even on the table.

The pattern is consistent: ownership structure determines what strategies are available. Go public too early and the market rewards short-term optimization. Stay private and you can spend 25-30% of revenue on marketing that won’t pay off for a decade.

But private ownership has real downsides too. There’s no access to public capital markets when you need a $1.7 billion facility - you fund it from cash flow or you don’t build it. There’s no stock-based compensation to attract top talent in an era where every major consumer company uses equity to recruit. And there’s no market discipline on capital allocation - what looks like “freedom” could just as easily be called “no accountability.” The Yoovidhya family holds the majority 51% stake from Thailand, and the actual mechanics of governance between two families on different continents remain opaque. Private doesn’t automatically mean better. It means different tradeoffs.

The $1.7 Billion Bet

There’s one more piece to the puzzle - and it’s the strongest evidence for the private ownership thesis.

In September 2025, Red Bull broke ground on a $1.7 billion production and distribution facility in Concord, North Carolina - a joint venture with Rauch and Ball Corporation. The facility spans 2.36 million square feet, with capacity for up to 3 billion cans annually. Ball is building an 800,000-square-foot aluminum can plant on the same campus. Operations are expected to begin in 2028, full capacity by 2031.

For 38 years, Red Bull never owned factories. Production was outsourced to Rauch. The company stayed asset-light so every dollar of structural investment went into the brand. Now it’s putting $1.7 billion into manufacturing infrastructure.

Try doing that as a public company. Try telling Wall Street you’re going to spend $1.7 billion - funded entirely from cash flow, not debt, not equity - on a facility that won’t hit full capacity for six years. Try explaining that you’re pivoting a 38-year-old asset-light model because you see a strategic opportunity in the US market, where Celsius and health-conscious challengers are growing fastest.

Red Bull is still partnering with Rauch and Ball - not building in-house capabilities from scratch. And it already tested the approach with a smaller joint facility in Glendale, Arizona in 2021 (around $140 million). Concord is the scale-up. But the point stands: only a private company with enormous accumulated cash flow could make this bet on its own terms.

The Risk

Two engines, private ownership, and now a $1.7 billion bet on the future. The case is strong. But this is where I want to pump the brakes and look at what could go wrong.

The health-conscious shift.

Celsius didn’t capture 10-16% of the US market by accident. Consumers, especially younger ones, are moving toward “better-for-you” energy.

• Celsius positions as a fitness supplement.

• Ghost targets lifestyle and gaming.

• Alani Nu speaks to wellness.

• Liquid Death announced plans to enter energy drinks in July 2025, with retail launch in early 2026, bringing lower caffeine and an irreverent health-adjacent vibe.

Monster and Red Bull are losing market share in an industrywide shift toward health and wellness, with retailers allocating more shelf space to these newer brands.

But it’s the cautionary tales that matter most. Prime Energy, built on Logan Paul and KSI’s massive owned media following (40M-plus YouTube subscribers), hit $1.2 billion in sales in 2023 using a version of the “brand-as-product” playbook.

By late 2025, revenue was projected to crash 76% to roughly $300 million. UK sales fell 48% in 2024 alone. When brand relevance fades, and the physical product is a commodity, there’s no fallback. Red Bull has avoided this for 35 years - but that’s execution excellence, not proof the strategy is inherently safe.

Red Bull’s response has been limited: Sugar Free, Zero, and an expanding Editions lineup. There is no Red Bull fitness sub-brand. There is no Red Bull wellness play. The brand identity is tied to extreme sports and adrenaline - and there’s a real question about whether that identity can stretch to meet health-conscious consumers where they are.

Culture risk

This arguably matters more than the other risks combined.

Helmut Marko, Red Bull’s longtime motorsport advisor, left at the end of 2025 after more than two decades. His parting words were pointed: “I felt something had been lost.” The internal dynamics are shifting - Red Bull’s leadership, with the backing of both the Mateschitz and Yoovidhya families, removed Christian Horner as F1 team principal. Marko publicly disagreed.

The succession team (Watzlawick, Kirchmayr, Mintzlaff) operates very differently from the singular vision of a contrarian founder. Mark Mateschitz deliberately stepped back from operations, saying: “I do not believe one should be both an employee and a shareholder of the same company.”

The question is whether a committee can sustain the kind of weird, contrarian, long-horizon decisions that a single founder made instinctively. Private ownership gives you the freedom to invest for the long term. It doesn’t guarantee that you will.

The US market erosion

Red Bull’s US volume share has been gradually declining as emerging brands gain ground, with combined Red Bull/Monster share falling from about 76% to 66%. Revenue keeps growing because premium pricing absorbs some volume loss - but the trend is directionally concerning.

What It All Means

Most companies treat marketing as a cost. An expense line to be trimmed, a percentage to be minimized. Red Bull treats marketing as the product. The drink costs $0.09. The brand costs $3 billion a year. The brand IS the manufacturing.

But here’s what the Red Bull story actually teaches you, and it’s not “spend more on marketing.” It’s that the structure of ownership determines the ceiling of strategy.

Two engines power Red Bull: marketing that builds appreciating assets and pricing that funds reinvestment. But both engines only run because private ownership removes the constraints that would shut them down. Go public, and Wall Street caps your marketing spend at 7-12% of revenue. Take on activist shareholders, and five-year bets become quarterly optimizations. Accept outside capital, and someone else’s time horizon replaces your own.

Here’s the diagnostic: when you see a company with patient ownership, strong reinvestment capacity, and the willingness to make long-horizon bets, you’re looking at a strategic ceiling that competitors can’t reach by simply copying the playbook - they’d have to replicate the ownership structure first. When ownership doesn’t match the strategy’s time horizon, something breaks.

• Monster has the margins but its public ownership caps how aggressively it can reinvest.

• Celsius got distribution through PepsiCo but public market pressure punished it for slowing growth even when revenue was still rising.

• Prime Energy had massive brand awareness but nothing structural underneath - when the brand faded, nothing remained.

Stripe, Databricks, and SpaceX figured out the same thing from different industries: stay private long enough to build structural advantages that public market pressure would prevent. Red Bull has done it longer than any of them - 35 years and counting.

The open question is whether the ceiling holds without the founder. Mateschitz set the entire structure in motion. His successors inherited it. But structures don’t maintain themselves. The health-conscious shift could erode the brand’s cultural relevance. A committee-led culture could gradually lower the ambition that private ownership was designed to protect.

Mateschitz said it himself: “Can you imagine me in a shareholders’ meeting?”

No. And that’s the point. The question now is whether the people running Red Bull can sustain the structural advantages that made that question rhetorical.

TLDR

→ Rain is a Visa Principal Member, the same direct relationship as banks like JPMorgan Chase, achieved by a company founded in 2021

→ They settle transactions with Visa 7 days a week using stablecoins, the first to ever settle on a weekend or holiday

→ $3B+ annualized TPV, 30x card growth, 200+ partners in the last year

→ The thesis: stablecoins become invisible infrastructure. Users see “dollar balance.” Merchants see “Visa transaction.” The blockchain never surfaces.

You can ask an AI to draft a contract and have it back before you finish blinking. You can stream video from a server 3,000 miles away without buffering. You can send money to the other side of the planet in—

Actually, no. You can’t.

Wire $10,000 on a Friday afternoon. Wait until Tuesday. The most sophisticated payment network on Earth still can’t move money on a Saturday.

Then, a four-year-old company settled with Visa on a Sunday, something that had never been done before.

Meanwhile, a freelancer in Argentina checks her app. $847.32. She taps her card at a café. Transaction approved. If you asked her what a stablecoin was, she’d shrug. She just knows the money works.

That gap, between a historic first and a woman buying coffee without thinking about it, is the whole story. To understand why it matters, you need to understand how money actually moves.

How Money Actually Moves

Here’s something most people never think about: when you tap your card, couple of critical steps happen.

One is authorization. Your card talks to the merchant’s terminal, which talks to the payment network, which checks if you have the funds. This happens in milliseconds. Visa’s network can handle 65,000 of these per second, anywhere in the world, any time of day.

The other is settlement. The actual movement of money from the bank that issued your card to the merchant’s bank. This is where things get weird.

Settlement doesn’t happen in milliseconds. It happens in days. Your bank doesn’t wire money to the merchant’s bank every time you buy a coffee. Instead, all the transactions get batched up and processed through a chain of intermediary banks, each with their own systems, their own time zones, their own operating hours.

Authorize a transaction on Friday afternoon. The merchant’s bank might not see that money until Tuesday. Why? Because settlement infrastructure runs on bank hours. Federal wire systems shut down at 6pm. Correspondent banks close for the weekend. A holiday in Singapore creates a gap that ripples through to New York.

This creates a problem: during that gap, someone is on the hook. The money has been authorized but hasn’t arrived. That’s credit risk. To cover it, everyone in the chain has to hold reserves, capital sitting idle, just in case.

The most powerful payment network on the planet, processing trillions of dollars annually, still builds credit risk into weekend gaps because the underlying money can’t actually move.

That’s the absurdity. Authorization lives in 2026. Settlement lives in 1975.

The Bridge

Stablecoins fix this.

A stablecoin is just a digital dollar that lives on a blockchain. The blockchain doesn’t have bank hours. It doesn’t take holidays. It doesn’t close at 6pm or wait for correspondent banks to open in different time zones.

When you settle in stablecoins, the money moves when the transaction happens. Not three days later. Not after a chain of intermediaries process their batches. Now.

This isn’t a crypto pitch. It’s a plumbing upgrade. Stablecoins are better rails.

And the market has noticed. Stablecoins crossed $300 billion in market cap this year. Visa and Mastercard have both integrated stablecoin settlement. The GENIUS Act gave them a federal framework in the U.S. Circle and Paxos got banking charters. The infrastructure layer is being defined right now—and the question is who builds it.

Enter Rain

Farooq Malik wasn’t a crypto evangelist. He was CIO at the North American Development Bank, watching settlement break constantly.

Every day, a team would sit there monitoring emails, trying to reconcile incoming wires with the transactions they represented. High velocity of payments. Low data fidelity. Money would arrive and nobody could figure out what it was for. Money would be sent and disappear into the correspondent banking void for days.

His co-founder, Charles Naut, came from a different world, Stanford CS, building a company that Intuit acquired, then principal engineer at QuickBooks. He’d built infrastructure at scale.

They were working on a side project together: a document generation tool for angel investments. Simple fintech plumbing, checkout link on top of safe notes, connected to ACH and wire rails.

Then a customer asked if they could pay in USDC.

The difference was an order of magnitude. ACH transfers were taking 7 to 10 days to arrive. Customers were emailing asking if their money had been stolen. The stablecoin payment? Instantaneous. Transaction finalized. Instant reconciliation. No team monitoring emails. No 7-day float.

That was the moment. Stablecoins weren’t a speculative asset. They were just better infrastructure.

The side project became Rain.

Subscribe now

What Rain Built

Here’s what it looks like in practice:

• A fintech in São Paulo integrates Rain’s API.

• Within weeks, their users can deposit reais (BRL, Brazil’s local currency), hold a dollar balance, and spend anywhere Visa is accepted, 150 million merchants, 150+ countries.

• The fintech didn’t negotiate with a bank. Didn’t find a BIN sponsor. Didn’t stitch together five vendors for compliance, KYC, and card fulfillment.

One integration.

In plain terms, Rain gives you three things:

• Money-in. Users deposit in any currency, convert to stablecoins. Local rails in, dollars out.

• Accounts. Embedded wallets that hold stablecoin balances. The user sees dollars; the infrastructure is on-chain.

• Money-out. Spend via card anywhere Visa works, or convert back to local currency. Stablecoins out, local rails out.

This is enabled by:

Direct relationship with Visa

Rain is a Visa Principal Member, the same status as JPMorgan Chase, Bank of America, the giants. Most fintechs work through intermediaries: you partner with a bank, who sponsors your program, who settles with Visa. Each layer adds friction, counterparty risk, and capital requirements. Rain skipped the layers. They settle directly with Visa. Their name on the back of the card. Their BIN. Their settlement stack.

For a company founded in 2021, without a bank charter, this is almost unheard of. Principal Member status typically takes decades of operating history. Rain got there in under four years.

7-day stablecoin settlement

Rain was the first company globally to settle a transaction with Visa on a weekend. The first to settle on a holiday.

I need you to understand how wild this is. Visa has been around since 1958. Sixty-seven years. No one had ever settled with them on a Saturday. Rain did it, and then did it again on a Sunday, and then on Christmas.

This is part of a broader evolution inside Visa. Cuy Sheffield, Head of Crypto at Visa, describes what’s been happening in their treasury.

“Our treasury historically has been two-dimensional—we receive dollars and pay out one of the 27 currencies we support. Then you add a third dimension: it’s not just currencies, but the form factor. Do you want to pay us in traditional dollars via Fed wire? Do you want to pay us in USDC? And then the fourth dimension: which blockchain?”

Rain was one of the first partners to push this live, helping Visa figure out what 4D treasury operations actually look like in practice.

The practical impact is capital efficiency. Traditional card programs need to prefund settlement:

• if you’re processing $100 million monthly and settlement takes 3-5 business days, you might need $15-20 million sitting in reserve to cover the float.

• At a 4% cost of capital, that’s $600-800K per year in dead money.

Rain’s just-in-time settlement drops that prefunding requirement to a fraction. The capital that would be sitting idle is now working. For a fintech running tight margins, that’s the difference between scaling and stalling.

Token and chain agnostic

Rain supports whatever stablecoin their customers need. Native deployments on Ethereum, Solana, Tron, and a dozen others. They’re not issuing their own token. They’re not betting on which blockchain wins. More on why this matters in a moment.

Full-stack simplification

Beyond settlement, Rain handles the operational complexity that makes card programs a nightmare to run yourself. KYC, compliance, transaction monitoring, card fulfillment. One API instead of stitching together five vendors.

The Numbers

The growth metrics are kind of crazy.

In the last twelve months:

• 30x increase in active cards

• 38x growth in payment volume

• Over 200 partners now running on the platform

• $3 billion in annualized TPV.

To put that in perspective: most card-issuing platforms take five to seven years to hit meaningful volume. Rain crossed $3 billion in their fourth year, and they’re growing 38x year-over-year. The compounding hasn’t even started.

The fundraising tells the same story.

• Series A was about proving the model: scaling card issuance and launching daily on-chain settlement with Visa

• Series B expanded the stack: onramps, virtual accounts, embedded wallets, rewards. A full suite so partners could build on one integration.

They just landed their Series C in January 2026: $250 million at a $1.95 billion valuation. That’s a 17x increase in 10 months. Investors backing the round keep saying the same thing: “This is one of the fastest-growing companies in our portfolio.”

When you’re the connecting layer for a $300 billion market heading toward $2 trillion, the numbers start to make sense.

The Invisible Infrastructure Thesis

Here’s what Rain figured out before everyone else: the stablecoins that win won’t be the ones users think about.

In Latin America, a phrase is emerging: “dollar card.” Not “stablecoin card.” Not “crypto card.” Dollar card. Top-of-wallet for daily spend.

• What the user sees: a dollar balance on an app.

• What the merchant sees: a normal Visa transaction.

• What neither of them sees: stablecoin settlement, blockchain rails

If you believe that’s the future, stablecoins as invisible rails, then Rain’s token and chain agnosticism makes perfect sense.

The stablecoin market is a land grab right now. Tether vs Circle vs PayPal vs whoever launches next. Ethereum vs Solana vs Tron vs whatever chain emerges. Consumer wallets fighting for users. Exchanges fighting for volume. Everyone betting on which gold strike pays off.

Rain isn’t betting on any of them. They’re selling the shovels.

USDC, USDT, PYUSD, Western Union’s new USDPT—Rain doesn’t care which wins. They support all of them. Ethereum, Solana, Tron, twelve-plus other chains—native deployments on each. The more tokens proliferate, the more chains launch, the more Rain benefits. They’re not trying to pick winners. They’re building the layer that connects all of them to spending.

That’s the thesis. Not “stablecoins win.” Not “Solana wins.”

The thesis is: stablecoins become invisible infrastructure, and whoever builds the layer that connects them to the real economy captures the value.

This matters because of where adoption is actually happening. In Argentina and Venezuela, over 30% of digital wallets now hold stablecoins for daily spending. Sub-Saharan Africa has the world’s highest stablecoin adoption rate. Remittances to Southeast Asia hit $18.6 billion in stablecoins in the first half of 2025.

These aren’t crypto enthusiasts. These are people in countries where local currencies lose 20%, 30%, 50% of their value annually. A dollar-denominated account isn’t speculative. It’s survival. And they have no idea there’s a blockchain involved. They just know the card works.

Rain isn’t replacing other card programs. They’re replacing cash. Kiosk payouts. SWIFT transfers. The friction that kept people from accessing dollar-denominated financial services in the first place.

Value accrues to whoever builds that connecting layer. Not the consumer wallets. Not the exchanges. Not the tokens. But the infrastructure no one sees.

The Moat Question

Let’s be direct: Rain’s position is a head start, not a permanent fortress.

Visa’s stated goal is to make stablecoin settlement a default capability for fintechs. They’re already expanding access, Lead Bank, Cross River, and more banking partners are coming online. The door Rain walked through is opening wider.

Rain’s advantage shifts from “only one who can do this” to “best at executing this.” That’s a different kind of moat.

But execution at scale is its own moat. 200+ partners. $3 billion in annualized volume. Every new integration, every new partner, every additional dollar of volume increases switching costs. Fintechs don’t rip out payment infrastructure lightly. The head start compounds.

The 7-day settlement capability will become table stakes eventually.

Rain’s bet is that by then, they’ll have enough partners locked in and enough volume flowing through that the structural advantage won’t matter. Series C, $250 million at a $1.95 billion valuation, a 17x increase in 10 months, gives them the capital to execute.

The obvious risks exist:

regulatory uncertainty as frameworks mature, heavy dependence on the Visa relationship, and exposure to stablecoin issuer failures. Rain’s multi-chain, multi-token approach mitigates some of this. But the bet requires the stablecoin ecosystem to keep maturing.

Where This Goes

The Western Union partnership tells you the trajectory. A 174-year-old remittance company, 400,000+ retail locations worldwide, just launched their own stablecoin and picked Rain as the first infrastructure partner for their Digital Asset Network. Legacy meets new rails. Stablecoin-to-cash conversion at the corner store.

Rain’s priorities are explicit: integrate recent acquisitions that deepen their routing and compliance stack, expand geographic reach, and launch products that make stablecoins feel like any other payment.

The pitch to traditional fintech isn’t “replace your stack with crypto.” It’s “upgrade your capital efficiency with better settlement infrastructure.” What would you build if your reserve requirements dropped by half?

The Infrastructure No One Sees

The best infrastructure is one you never think about.

Rain bet on that thesis before most people understood it, building settlement capabilities and capital efficiency tools while others were still debating whether stablecoins were real. Four years in, with $3 billion flowing through their systems, it’s looking like they timed it right.

If they are right about where this goes, we’re watching something bigger than one company’s success. The settlement layer that banks have controlled for decades is migrating to new rails. The next generation of fintech founders won’t start by finding a BaaS partner and negotiating for a BIN. They’ll plug into infrastructure like Rain and ship globally from day one.

A freelancer in Argentina checks her app. $847.32. She taps her card at a cafe. Transaction approved.

She has no idea any of this that is happening underneath. She’s just buying coffee.

That’s exactly how money should be.

Subscribe now

Key Insights

→ Brutal efficiency: Bending Spoons acquires fading tech brands, cuts 70-80% of staff, and runs them better than the original teams. A playbook now applied to Evernote, Vimeo, and AOL.

→ The secret ingredient: Bending Spoons built 50+ proprietary tools and a 0.04% hire rate that lets 500 people run what would normally require thousands. A technical moat PE firms can’t replicate.

→ The AI accelerant: What made this model hard to copy is becoming easier as AI coding tools compress the talent needed to run software at scale.

→ IPO on the horizon: Backed by T. Rowe Price and Fidelity, a 2026 listing could be one of Europe’s largest tech IPOs.

In March 2025, employees at Komoot, the beloved route-planning app for cyclists and hikers, logged into their systems to find their company had been sold. The buyer was Bending Spoons, an Italian firm with a reputation that preceded it. Within two weeks, 85% of the 150-person team was gone.

One longtime employee described it to Bikepacking.com as “a rough and cruel betrayal.” The founders walked away with the lion’s share of nearly €300 million. The staff who dedicated up to 15 years to building the product? Most got severance packages and a goodbye video.

This is the Bending Spoons playbook: buy fading internet brands, cut 70-80% of the staff, plug everything into a shared platform, run it better than the original team ever did. It sounds brutal because it is. But here’s the uncomfortable truth:

It works. And in the age of AI, it’s about to become a template.

Why This Matters Now

The cuts Bending Spoons made to WeTransfer and Evernote happened with 2023-era AI tools. GPT-4 could write code and answer questions, but it couldn’t reliably ship features or debug production systems.

That’s changing fast as the frontier models keep pushing the benchmark forward. Agentic systems are starting to handle multi-step workflows: gathering requirements, writing code, running tests, and deploying changes. Bending Spoons’ obsession with modern tech stacks during transformation isn’t accidental. Newer code works better with AI generation tools.

Luca Ferrari, Bending Spoons’ CEO, told Patrick O’Shaughnessy on Invest Like the Best that AI is “definitely impacting” how they think about portfolio operations. The implication is straightforward: the same 500 people can run more products. Or the same products can run with fewer people.

Bending Spoons’ model has been hard to replicate because it requires a specific kind of technical organization. 500 exceptional engineers who can absorb and maintain a dozen different products, plus proprietary tools that took years to build. Most acquirers don’t have that.

But AI changes the skill curve. A smaller team with better tools can now do what Bending Spoons does. The playbook that seemed like an outlier starts to look like a template.

The $40,000 That Started Everything

Let’s rewind to how it all began for Bending Spoons.

In 2010, five engineering students from Italy were living in Copenhagen, trying to build a startup. Luca Ferrari had grown up in a tiny village in northeast Italy. “Shy, weird, gentle,” as he later described himself on the 20VC podcast. Not the obvious founder type.

They built a photo-sharing app called Evertale, pitched it at TechCrunch Disrupt SF 2011, raised seed funding, pivoted, and then watched it all fall apart. Most founding teams scatter after a death like that. These five made an unusual pact: whoever could get a paying job would share their salary with the others.

Ferrari got the best offer, a position at McKinsey. For about a year and three months, he lived a split life: management consultant by day, sending his salary to his broke co-founders to cover their rent and food. When Evertale finally shut down, the VCs didn’t want to deal with liquidation paperwork. They sold their shares back for $1, leaving the team with about $40,000.

That $40,000 became the seed for Bending Spoons.

The failure taught Ferrari something important: they had been arrogant. “We were arrogant in thinking that we knew what the market wanted,” he explained. They’d spent years building a product based on assumptions. They were wrong.

The insight that followed was counterintuitive. Ferrari realized startups have two phases. The zero-to-one phase is chaotic: figuring out what to build, who to sell it to, how to make it work. This phase relies on luck, timing, and things outside your control. But once a company figures that out, it enters the one-to-n phase: scaling what works, reaching more people.

The Bending Spoons thesis: outsource the zero-to-one risk entirely. Let other founders take those bets. Then, when a company has proven product-market fit but hit a plateau, swoop in. The one-to-n phase is based on skill—engineering, operational excellence, disciplined execution. And those are things you can systematize.

Subscribe now

The Machine

The first deal was tiny: a keyboard customization app for iOS. Price tag: around $10,000. Buy it, improve it, reinvest the profits. Then do it again. Small apps. Photo editors. Utility tools. The kind of unsexy products that rack up downloads without making headlines.

What emerged was a self-reinforcing flywheel.

1. Acquire businesses that have hit a ceiling.

2. Transform them by plugging into a proprietary platform.

3. Generate excess cash flow.

4. Reinvest into the platform and bigger targets.

5. Repeat.

By 2022, the company was generating over $100 million in annual revenue with more than 500 million app downloads. And it had never raised significant equity capital, for almost a decade, Bending Spoons bootstrapped its growth using retained earnings and debt.

Then came the deals that changed the scale.

Evernote in November 2022—the once-beloved note-taking app that had fallen from its $1 billion valuation peak. WeTransfer in July 2024. Vimeo for $1.38 billion in September 2025. AOL for ~$1.5 billion in October 2025. Eventbrite for $500 million in December 2025. All told, Bending Spoons now manages a portfolio of internet brands with over 400 million monthly active users.

The Transformation Playbook

Buying the asset is just the prologue. Bending Spoons breaks transformation into two phases.

Phase One: The Platonic Ideal. Right after acquisition, Bending Spoons embeds product managers and engineers to deeply understand the product, the customers, and how everything works. The goal: what is the best possible version of this product?

Phase Two: Execution. Technology gets modernized. Product velocity increases—Evernote went from shipping updates every four to eight weeks to shipping weekly. Pricing gets raised, often dramatically. And teams get restructured, which is the polite way of saying most people get fired.

What makes this work isn’t just operational discipline. It’s proprietary infrastructure. Bending Spoons has built roughly 50 internal technologies that accelerate every transformation.

• Pico is the analytics platform, hundreds of thousands of queries daily, A/B testing everything from UI changes to paywall designs.

• Manurva predicts customer lifetime value years into the future.

• Pantheon handles ML ops across all products.

Think of it like Palantir’s forward-deployed engineering model, but applied internally across portfolio companies. Multiple SWAT teams that can be deployed wherever the highest-ROI opportunity exists.

The 0.04% Filter

Bending Spoons received over 700,000 job applications in 2025. They hired about 100 people. That’s a 0.04% acceptance rate, more selective than any elite university on the planet.

The company primarily hires recent graduates. The logic: they don’t come with bad habits from other companies. According to Sifted, a 28-year-old named Florinda Pannofino who joined as a marketing trainee while still at university now leads the entire Evernote business unit, a team pulling in €200 million a year. “I kinda grew up here,” she says.

Based in Milan, Bending Spoons has a near-monopoly on top technical talent in Europe. Graduates from Europe’s best universities are exceptionally capable but don’t have the same density of opportunities they’d find in Silicon Valley. Bending Spoons captures that arbitrage.

This selectivity creates a moat. Running a dozen products with 500 people requires each of those people to be exceptionally capable. You can’t build that kind of organization overnight.

The Human Cost

Let’s be direct: this efficiency has a price.

When Bending Spoons acquired WeTransfer, they announced plans to cut 75% of the 350-person staff. When Komoot was acquired, “employees signed in one morning to learn that the company many had dedicated up to 15 years of their life to was no more,” according to Bikepacking.com. “Eighty percent were fired immediately.”

When Mosaic Group’s assets were purchased in January 2024, all 330 employees were laid off. They weren’t even part of the acquisition. FiLMiC’s entire original team was let go in late 2023.

Ferrari has been transparent about the logic. The company “studies the organization” after each acquisition and “determines what is the best vision of the business going forward.” That vision almost always involves far fewer people.

There’s a defensible business case. Many of these companies were struggling. Evernote was in decline for years before the acquisition. WeTransfer had failed to IPO in 2022 due to market conditions. The alternative might have been slow death rather than efficient operation.

But the human math is still hard to stomach. €190 million in restructuring costs in 2025, that’s severance packages, transition costs, and the human friction of rapid integration. $2.1 million in revenue per employee is an impressive efficiency metric. It’s also hundreds of people learning their jobs are gone overnight.

Where It Breaks

Not every acquisition is a success story. FiLMiC offers a cautionary tale.

Bending Spoons acquired the professional video app in September 2022 for €21.5 million. FiLMiC served a small but passionate community of mobile filmmakers, power users who took their craft seriously.

The standard playbook was applied: shift from one-time purchases to weekly subscriptions, raise prices. What worked for casual user bases didn’t work for FiLMiC’s prosumers. The community revolted. According to Sensor Tower data, FiLMiC’s revenue has been in steady decline ever since.

The lesson: the model works best with casual users who are locked in by accumulated data (like Evernote notes) or habitual use. Power users with specific needs can reject the changes and leave.

The Numbers

Between 2018 and the first half of 2025, Bending Spoons’ revenue grew more than 40-fold, from €30 million to over €1 billion. With the Vimeo, AOL, and Eventbrite deals, that figure is set to nearly double again.

But the headline isn’t revenue. It’s what they keep. Bending Spoons runs at roughly 50% adjusted EBITDA margins. That’s extraordinary for a software company managing this many products. Most SaaS businesses would be thrilled with 25%.

The capital structure reflects the ambition. In October 2025, Bending Spoons raised $710 million in equity at an $11 billion pre-money valuation. T. Rowe Price led the round, with Baillie Gifford, Fidelity, and Cox Enterprises participating. The company also secured a $2.8 billion debt package from J.P. Morgan, Goldman Sachs, Wells Fargo, and others.

Early backer Tamburi Investment Partners is sitting on a 20x return.

Why Traditional PE Can’t Compete

Private equity firms could, in theory, do what Bending Spoons does. Buy struggling tech companies. Cut costs aggressively. Optimize for profitability. In practice, they can’t. The gap is structural.

PE firms are financial engineers, not software engineers. They can optimize pricing and renegotiate contracts. They can’t rebuild a product’s backend, migrate it to new infrastructure, or deploy Manurva to predict LTV three years out. That requires a permanent, deeply technical organization—the kind of team that doesn’t exist in a fund structure.

The timeline matters too. Deep technical integrations take years. A PE firm with a 4-year exit horizon can’t justify investments that mature in year 6. Bending Spoons’ “hold forever” model lets them make long-term bets that traditional PE economics don’t support.

Ferrari describes the company as “25% PE, 75% tech company.” That framing isn’t marketing, it’s a competitive moat.

The Endgame

Bending Spoons is approaching a threshold that typically triggers an IPO: $1 billion in EBITDA, backed by blue-chip institutional investors. Analysts expect a listing in 2026—likely one of Europe’s largest tech IPOs.

Ferrari has said they’re “trying to build one of the most successful companies of all time.” On Invest Like the Best, he described himself as “perennially unhappy”, a mindset that fuels relentless improvement and defines the culture at Bending Spoons.

The question isn’t whether the model works. €1 billion in EBITDA, €13 billion in enterprise value, and 20x returns for early investors suggest it does. The question is whether Bending Spoons remains an outlier or becomes a category.

PE firms are watching. Some are building internal tech capabilities. Others are partnering with operating companies that can handle the integration work. And as AI continues compressing the headcount required to run software products, the talent advantage Bending Spoons built over a decade could become accessible to anyone with the right tools, and the stomach for the cuts.

There’s something uncomfortable about celebrating a company whose core competency involves firing most of the people who built the products they buy. But there’s also something clarifying about it.

For decades, tech operated under the assumption that growth required growth, more users meant more engineers meant more managers meant more complexity. Bending Spoons flips that assumption. A smaller, more talented team can accomplish more. Shared infrastructure beats redundant effort. Long-term ownership beats the exit treadmill.

Whether you find this inspiring or dystopian probably says more about your priors than about Bending Spoons. They’re not hiding what they do. Sellers know what’s coming. The model is transparent, even if it’s brutal.

The employees at the next acquired company already know the playbook. Somewhere right now, a team is building something they love, unaware that a company in Milan is reviewing their metrics, running the numbers, and preparing an offer.

That offer will be hard to refuse. It always is.

Subscribe now

Here’s the Answer

Researching is a core part of my work. Regulatory and licensing questions come up regularly, and getting them wrong has real consequences.

When I asked Google, I got ads for compliance consulting services, a legal blog from 2022, and several articles that discussed the laws generally but never specifically.

But Perplexity gave me a detailed answer in eight seconds, citing updated regulations, a fintech industry analysis, and specific licensing threshold changes. I could immediately verify everything against the source documents.

That's when Perplexity clicked for me. It's not just faster than Google, it's fundamentally different. Instead of giving me homework, it does the research and shows its work through citations.

This experience explains why 30 million monthly users have made Perplexity part of their workflow, and why the company just raised $100 million at a $18 billion valuation in July. But it also explains their core challenge: building something defensible when the underlying technology is increasingly commoditized.

The Founding Story: From Google Fan to Search Challenger

Aravind Srinivas had always been fascinated by Google. The 29-year-old AI researcher spent time at DeepMind and admired Sundar Pichai as a fellow Chennai native who'd made it to the top of tech. But by 2022, that admiration had evolved into something different: a conviction that he could build something better.

Traditional search was broken. Why click and sift when AI could read the web for you?

The timing turned out to be accidentally perfect. Knowledge workers were drowning in micro-questions—Slack threads full of "quick question" messages, Notion docs that needed one specific statistic, Figma comments asking for competitor context. The old ritual of opening a new tab, typing a search, scrolling past ads, and clicking through to sources felt increasingly clunky.

Google's results were getting noticeably worse, not because their algorithm broke, but because the economic incentives did. Every search is packed with more ads, more "People also ask" boxes, more featured snippets that answered adjacent questions. The "scroll to real info" tax was getting expensive for power users.

Srinivas left his research scientist role at OpenAI with a radical vision:

Instead of treating the web as a distant library, he wanted to treat it like RAM, instantly readable, available in real-time.

So in early 2022, he and co-founders Denis Yarats, Johnny Ho, and Andy Konwinski (seasoned crew from DeepMind, FAIR, and Berkeley) built a prototype and took it to Y Combinator's Winter '22 batch. The demo: AI that answers questions with cited sources, in plain language.

However, speed isn't a moat if GPUs are rentable. If your only advantage is doing what Google does but faster, remember that Google can simply throw money at NVIDIA to close the gap.

This tension, raw conviction versus the need for a moat, became Perplexity's defining challenge. By mid-2023, Perplexity's "direct memory access" dream was live. Users flocked. But the real work was just beginning.

The Core Hook That Worked

The foundation is deceptively simple: free search at perplexity.ai, augmented by Perplexity Pro at $20/month. You know the experience: type a question, get a cited answer in roughly 10 seconds. The UX feels like ChatGPT but with clickable footnotes that actually work

Under the hood, Perplexity does the Googling for you: retrieving web results, then summarizing via LLM magic. It's like a personalized Wikipedia entry generated on demand, plus follow-up suggestions to keep you exploring. The speed addiction is real. Complex multi-hop queries are resolved in 5-15 seconds.

Pro converts power users with faster responses, longer answers, and "Copilot" for conversational refinement. At $20/month—same as ChatGPT Plus—it targets the "premium unlimited" crowd. Even a 1% conversion of 30 million monthly active users (MAU) equals roughly 300,000 paying users, generating about $6 million monthly revenue.

But here's the uncomfortable truth: this is easy to clone.

The basic recipe is simple: search the web for relevant content, then feed that content to an AI model that writes a summary with citations. This approach, called "Retrieval-Augmented Generation" (RAG), is like having an AI research assistant that first gathers materials from the library, then writes you a report with footnotes.

By mid-2024, independent developers released Perplexity-like systems using OpenAI's APIs and Google's search tools. One GitHub repo, cheekily named "Perplexica," nearly carbon-copied the experience with a few hundred lines of code. Anyone with basic programming skills and $50/month in API credits could build something remarkably similar.

The tech isn't a moat. Perplexity's edge comes from integration finesse and proprietary tuning earned over time. But fundamentally, retrieval-augmented Q&A is becoming commodity. Which brings us to their first big bet on defensibility.

Subscribe now

The AI Browser Gambit: Building Moats Before Competitors Wake Up

Perplexity's boldest move was Comet, an AI-native browser they launched in early 2024, months before Google or Apple announced similar initiatives. This wasn't just feature expansion; it was a race to own distribution before the giants realized the threat.

Aravind calls it an "agentic browser", one that doesn't wait for clicks and typing but takes actions autonomously. The demo vision: tell your browser, "Find me a good Italian restaurant in SF and book a table for 2 at 7pm tomorrow." Instead of you navigating OpenTable, Comet's AI agent searches for well-rated spots, checks availability, and actually books the reservation, filling forms, handling confirmations.

The strategy was pure first-mover advantage. If Perplexity could establish Comet as the go-to AI browser before Apple Intelligence or Google's Project Astra matured, they might lock in distribution advantages that would take competitors years to overcome.

Still, the timing calculation made sense. Perplexity launched Comet when the giants were still treating AI as a feature, not a platform. Now that Apple and Google are racing to integrate AI into their browsers, Perplexity's early position in agentic browsing might be their most valuable asset.

The Enterprise Table Stakes

While consumer products grab headlines, Perplexity needed an enterprise offering to compete with ChatGPT and Claude, both of which now offer private data access for businesses. Think "Perplexity inside your firewall", companies feed internal knowledge bases into the system, and employees ask questions like "Where's the latest pricing strategy deck?" getting instant answers with citations from confidential files.

They launched Enterprise API and Perplexity for Business in late 2023, connecting to Google Drive, Confluence, and databases. Natural language querying replaces SharePoint search hell—a value proposition that enterprise buyers understand.

But this isn't really a differentiator anymore. ChatGPT Enterprise and Claude for Work offer similar capabilities, often with better model performance and broader feature sets. Perplexity's enterprise play is more about staying competitive than building moats, they needed to check this box to avoid being eliminated from enterprise evaluations.

The enterprise market does offer stickier revenue than consumer subscriptions, and Perplexity's search-specific focus might appeal to companies that prefer specialized tools over general-purpose AI assistants. But it's increasingly a commodity market where model quality and pricing matter more than specialized positioning.

Distribution: The Bundling Strategy

Here's where Perplexity shows genuine strategic insight. Instead of fighting Google head-on for search market share, they're pursuing distribution through partnerships and bundling, getting Perplexity into users' hands as value-added services.

Financial Services Bundling: Revolut offered Perplexity Pro as a perk to premium banking customers in 2024, seeding thousands of new users who wouldn't have paid otherwise. The bank gets a differentiated offering; Perplexity gets exposure to high-value customers who might convert long-term.

Telecommunications Partnerships: Several mobile carriers internationally are bundling Perplexity subscriptions with premium data plans. The logic is simple: carriers want to differentiate beyond speed and coverage, and AI-powered search feels cutting-edge enough to justify plan upgrades.

OEM Deals: Perplexity secured a partnership with Motorola to integrate their AI search into select smartphone models, a small but meaningful win that proves the distribution strategy can work. They're also reportedly in discussions with Samsung and Apple for broader integration opportunities.

It's the Google playbook (recall Google pays Apple $15+ billion yearly to remain iPhone's default search). Perplexity can't throw that cash around, but they can offer smaller OEMs "AI search" differentiation that helps justify premium device pricing, while potentially providing larger manufacturers with innovative features that distinguish their AI capabilities.

This distribution strategy acknowledges a hard truth: most users won't voluntarily switch from Google. But they might use Perplexity if it comes "free" with something they already value.

The Economics of Instant Answers

Perplexity's business model is both promising and precarious.

Start with the good news: Perplexity had cracked something that most AI companies struggle with, actual paying customers. With 260,000 people paying $20 monthly out of 1.6 million subscribers, they'd achieved a 16% conversion rate. For context, ChatGPT, with 500 million weekly users, only converts about 4% to paid plans. That's genuinely remarkable.

But here's where it gets complicated. The company generated $59 million in revenue but $27 million was spent (almost half) went to discounts and promotions, meaning they're essentially paying people to use their product through various distribution channels.

Perplexity claims healthy 57% gross margins, but there's an accounting sleight of hand here. They classify $33 million spent serving free users as "R&D" rather than cost of revenue, while companies like OpenAI count similar expenses as direct costs. It makes the margins look rosier than reality.

The fundamental problem is structural. Monthly expenses are expected to hit $12 million by August—20% higher than they forecasted. Every new user makes this worse, not better. When Google serves search results, it costs essentially nothing because they're pulling from pre-built indexes. When Perplexity answers your question, they're running expensive real-time computations while paying other companies for both the AI models and web data.

Perplexity is betting that AI will become so cheap that real-time answer generation becomes economically viable at massive scale. It might work. Or they might discover that some services are inherently expensive to deliver, no matter how much technology improves.

Google's Innovation Dilemma and the AI Awakening

Google's initial response to Perplexity reveals everything about why startups sometimes beat incumbents, even with inferior resources.

The Innovator's Dilemma in Action When Perplexity launched, Google could have easily built similar citation-heavy answers. They had better data, more AI talent, and infinite compute budget. But they faced a classic disruption trap: their existing business model couldn't support the new approach.

Google makes $150+ billion annually from search ads. Every time they provide a direct answer instead of a list of links, they potentially reduce ad clicks. Features like AI summaries create internal tension between user experience and revenue optimization.

Meanwhile, regulatory scrutiny meant Google couldn't radically overhaul search UX without triggering new antitrust concerns. Publishers were already suing over featured snippets. More aggressive AI answers would intensify those battles.

The Awakening But Google wasn't asleep—they were strategically constrained. By mid-2023, they launched Search Generative Experience (SGE), essentially their version of Perplexity's cited answers. The quality was comparable, sometimes better, but the integration felt half-hearted. AI summaries appeared only for some queries, often buried below traditional results and ads.

Now Google is accelerating. Project Astra promises more conversational search. Gemini integration makes AI answers more prominent. They're finally treating AI as platform, not feature, but they had to overcome organizational antibodies first.

The Competitive Reality Google's innovation dilemma gave Perplexity an 18-month head start to build user habits and brand recognition. But that window is closing. Google has superior data moats (owning YouTube, Gmail, Chrome browsing data), better AI infrastructure, and obviously better distribution.

The question isn't whether Google can build Perplexity-like features. They already have. The question is whether Perplexity can leverage their head start into durable competitive advantages before Google fully commits to cannibalizing their own business.

The Competitive Landscape

The other major threat comes from AI-first companies that own the underlying models powering the entire ecosystem.

OpenAI: The Platform Play OpenAI's ChatGPT added web browsing capabilities by late 2023, directly competing with Perplexity's core value proposition. Their advantage: they own GPT-4, the model that powers much of Perplexity's backend. They can offer similar functionality at lower marginal cost while funding development through ChatGPT subscriptions and API revenue.

OpenAI also has 100+ million monthly active users already trained to use conversational AI. If they perfect web search integration, why would users switch to a specialized tool?

More threatening still, OpenAI is rumored to be developing their own web browser, which would directly challenge Perplexity's Comet browser strategy. If OpenAI launches a ChatGPT-powered browser with native search integration, it could eliminate both of Perplexity's key differentiators, specialized search UX and browser distribution, in one move.

Anthropic: The Enterprise Angle Anthropic's Claude increasingly handles complex research tasks with web access. Their advantage: superior reasoning capabilities and growing enterprise adoption. Companies already paying for Claude might not need separate Perplexity subscriptions.

The Capital Reality Both OpenAI and Anthropic have raised billions specifically for model development and infrastructure. They can afford to price their web-enabled AI services aggressively to gain market share, potentially making Perplexity's business model unsustainable.

Perplexity's counter-argument: they're building the best user experience for search-specific tasks, while the AI giants are building general-purpose tools. Specialization might provide enough differentiation to justify premium pricing.

But this assumes users will pay for multiple AI tools rather than consolidating around one excellent general-purpose assistant. Early evidence suggests most users prefer simplicity over specialization.

The Future: Niche Player or Acquisition Target

Here's my honest take on where this heads, and why I think Perplexity faces an uphill battle for independent scale.

The Consolidation Thesis AI markets are likely to consolidate around a few dominant platforms rather than fragment across specialized tools. Most users won't subscribe to separate AI services for search, writing, coding, and analysis. They'll choose one primary AI assistant that handles everything reasonably well.

This dynamic favors companies with broad capabilities and deep pockets, OpenAI, Google, Microsoft, Anthropic, over point solutions like Perplexity. Even if Perplexity builds the best search experience, they're competing for share of wallet against platforms users already pay for.

The Distribution Disadvantage Google processes 8 billion searches daily with zero marginal acquisition cost. Apple controls default browser settings for a billion iPhones. Microsoft bundles AI search into Windows and Office. Perplexity needs users to actively choose their product over deeply integrated alternatives.

While bundling partnerships help, they don't solve the fundamental distribution asymmetry. Perplexity is fighting for attention against companies that own the operating systems, browsers, and platforms where most work happens.

The Acquisition Logic All this points toward two likely outcomes: Perplexity becomes a successful niche player serving specific power-user segments, or they get acquired by a larger platform seeking AI search capabilities.

Acquisition makes strategic sense for several potential buyers:

• Meta: Needs AI search to reduce Google dependence for Facebook/Instagram information needs

• Amazon: Could integrate Perplexity into Alexa for better question-answering

• Apple: Perfect fit to supercharge Siri with citation-based search capabilities across their ecosystem

• Microsoft: Hedge against OpenAI relationship risks

The Optimistic Scenario: The Google Memo is Right About Models

There's a strong bull case for Perplexity embedded in Google's "We Have No Moat" memo, but it requires reading between the lines.

The memo argues that foundation model advantages are ephemeral and expensive to maintain. Open source alternatives catch up quickly, making massive training runs poor investments.

This is potentially fantastic news for Perplexity, because they're not playing the foundation model game at all. While OpenAI and Anthropic burn billions training models from scratch, Perplexity post-trains existing models and competes on application layer innovation, citations, speed, search-specific UX.

If foundation models become commoditized, value shifts to building excellent products on top of those models. That's exactly Perplexity's strategy. They can adopt whatever open source models work best without legacy constraints, while companies like OpenAI must recoup massive training investments.

In this scenario, Perplexity's capital efficiency becomes a huge advantage. They can iterate faster, price more aggressively, and avoid the innovator's dilemma that traps companies protecting expensive model investments.

My Prediction I lean toward the consolidation scenario, but I'd love to be wrong. The AI market is moving too fast for definitive predictions, and Perplexity has consistently exceeded expectations.

Most likely outcome: Perplexity builds a sustainable business serving 20-30 million users who genuinely prefer their search experience. They generate $500 million - $1 billion annually from subscriptions and enterprise contracts. Eventually, they either accept acquisition from a strategic buyer or remain independent as a high-quality niche player.

Either outcome validates their approach, even if it doesn't achieve the Google-killer ambitions some investors envision.

What AI Founders Should Steal

Perplexity's journey offers crucial lessons for anyone building in the AI space, regardless of where they ultimately land.

Distribution Trumps Technology The best AI model in the world is worthless without a path to users. Think distribution from day one: bundling partnerships, integration opportunities, pre-install deals. Perplexity succeeded not because they had superior AI, but because they found creative ways to get in front of users. Pure technology plays rarely win without solving the go-to-market puzzle.

Focus Beats Features Resist the temptation to become an everything app. Perplexity succeeded by doing one thing exceptionally well, cited search answers, while competitors added kitchen-sink features. Specialization can beat generalization if you're significantly better at the core use case.

Model Commodity Defense Don't build a company that depends on exclusive access to superior AI models. Those advantages are temporary and expensive to maintain. Instead, build differentiation through user experience, data network effects, or distribution advantages that persist even when model capabilities converge. Perplexity's post-training approach proves you can win without owning foundation models.

The broader lesson: AI creates winner-take-most markets where a few platforms capture disproportionate value. Success requires either becoming one of those platforms or finding defensible niches within them. Perplexity chose the latter path, we'll soon learn whether it leads to sustainable independence or strategic acquisition.

Either way, they've proven that startups can still challenge tech giants by moving faster, focusing sharper, and building products users genuinely love. In an industry racing toward commoditization, those advantages matter more than ever.

I've sat through 16 hours of meetings this week. I remember about 20 minutes of them.

Not because the meetings were back-to-back chaos, they weren't. But because I spent most of them frantically scribbling notes instead of actually listening. You know the feeling: someone's explaining a key decision while you're still writing down the previous point, and suddenly you've missed the most important part.

The worst part isn't the bad notes. It's realizing you were physically present but mentally absent from your own meetings.

The Software 3.0 Moment

That attention split isn't just a personal failing. It's a fundamental mismatch between how we think and how technology has been designed to help us.

I was wrestling with this when I watched Andrej Karpathy's recent YC presentation.

The former Tesla AI director laid out an evolution that clicked immediately. Software 1.0 required programmers to spell out every rule. Software 2.0 let us show AI examples and have it figure out patterns. But Software 3.0? We just tell the AI what we want in plain English.

The progression is stark when you see it mapped out. Most AI meeting assistants represent Software 2.0, they're trained to transcribe and summarize, but they can't adapt to how you think about meetings.

Software 3.0 is different. Neural networks have now become programmable with large language models (aka using English). AI can reshape itself based on context and intent. Not chatbots following scripts, but intelligence that learns your specific workflow and adapts accordingly.

As I watched his presentation, I realized his presentation ties in neatly with the startup that I wanted to do a deep dive about.

The Curse of the Transcript

We've all sat there at 4:55 PM, staring at a wall of text from the last meeting while already late for the next one.

Microsoft says we now spend 57% of work time in meetings and communications. Look at your calendar - it's probably worse. The five-minute buffer between calls that used to let you process what happened? Gone.

What makes this particularly frustrating is the sheer uselessness of most meeting documentation. A 10-page transcript dumps everything that was said but captures nothing of what was meant:

• Those five minutes where someone couldn't figure out his microphone

• The tangent about weekend plans that ate 10% of the call

• The circular discussion that eventually reached consensus but only after repeating the same points six different ways

You don't need a record of words. You need meaning. Not conversation, but conclusions.

Subscribe now

Granola: The Quiet Revolution in My Menu Bar

Sitting in my menu bar is a tiny app that I love called Granola. It's deceptively simple: an AI notetaker that listens to your meetings and turns them into concise, useful summaries. No bots joining your calls, no complicated setup - just a small icon that captures what matters.

From Problem to Product

Granola records your device's audio output. No bot joins your call. No recording announcements. Just a quiet menu bar icon.

When the meeting ends, Granola automatically processes the conversation through AI models to generate structured notes with key points, decisions, and action items. It even integrates any notes you typed during the meeting into its final summary.

The Right Problem to Solve

Granola's journey was messier than the clean interface suggests. When looking at the explosion of AI capabilities, it wasn't immediately obvious which problems were best suited for foundation models.

The team experimented with various prototypes, from AI journaling apps to coding assistantsm searching for the right fit. In dozens of customer development conversations, one pain point kept surfacing above all others: the overwhelming burden of meetings.

"For people whose jobs revolve around meetings, the work just stacks up," noted Chris Pedregal, Granola's founder and a former Google product manager. The pandemic's shift to wall-to-wall Zooms only amplified this problem.

What made meetings particularly attractive as a beachhead was their clear structure - defined start and end times, plus built-in virality. When one person used a new meeting tool, everyone else in the call was exposed to it. By March 2023, they officially founded Granola with a mission to tame "meeting hell" with AI.

Ambient AI at Work

Karpathy's Autonomy Slider

In his recent YC presentation, Karpathy introduced a concept that perfectly frames what makes Granola different: the autonomy slider.

It's a practical framework for thinking about AI's role in our tools. On one end of Karpathy's slider sits fully manual operation where humans do all the work. On the other end sits complete automation where AI operates independently. The magic, Karpathy argues, happens not at either extreme but at carefully chosen points along this spectrum.

As he puts it, "It's less Iron Man robots, and more Iron Man suits that you want to build." The distinction is crucial - augmentation versus replacement.

Most AI meeting tools have positioned themselves awkwardly on this slider.

• Some remain too manual, requiring users to start recordings, tag important moments, and organize summaries themselves.

• Others attempt too much automation, making decisions about what's important without human input, often missing crucial context.

Granola found a fascinating balance point - what Karpathy would call "partial autonomy with verification." During meetings, it sits far to the autonomous end - recording and processing without requiring any input. But immediately after, it slides back toward the middle, producing summaries that humans can verify, modify, and share.

Granola’s UI is philosophically aligned with Karpathy's vision of the future of AI tools. As he noted in his talk, "We need to build partial autonomy products... with an autonomy slider that can be adjusted depending on the task."

Granola's Autonomy Slider in Action

What makes Granola particularly interesting is how it dynamically adjusts this autonomy level throughout the meeting lifecycle:

Before a Meeting: Maximum AI Autonomy

Granola quietly prepares in the background. It syncs with your calendar to pull in the meeting title, attendees, and any prep materials. This gives the AI a head start on context, it can recognize that the 10am is a sales call vs. a design review, or that “ACME Corp” is a key client name.

During a Meeting: Maximum AI Autonomy (in the background)

Granola’s interface is a split-screen editor that looks comfortingly familiar. A blank note canvas – akin to Apple Notes or Google Docs. This is deliberate; the team wanted zero learning curve. You type notes as you normally would, capturing key points or decisions in your own words. The twist is that you don’t have to write down everything. Granola’s AI is already “taking notes” in parallel by transcribing every word. You can focus on distilling insights or tagging important moments. Throughout the meeting, there’s no AI distraction or weird UI: just a notepad. If you ignore Granola completely and take no notes, it will still have your back (more on that in a moment), but the design gently encourages a human-led outline.

Immediately after the Meeting: Partial Autonomy with Human Verification

Granola enters its real magic “post-processing” phase, transforming raw conversation into structured intelligence. What you get isn't just a transcript with a few highlights. It's an organized summary that captures:

• Key decisions reached during the discussion

• Action items assigned (and to whom)

• Questions that remain open

• Quotes that deserve emphasis

Importantly, the AI-generated additions are distinguished in color and linkage. Granola renders the AI-written text in gray, below the user’s own text in black. Every AI-added bullet also carries a hyperlink; clicking it jumps to the exact point in the transcript (or recording) where that info came up.

This way, you can verify any content the AI contributed or get more context if something looks off. The result is a meeting note with a combination of your inputs and highlights, combined with the AI’s complete memory.

Beyond the Meeting: Collaborative Knowledge Management

Granola's newest features extend beyond individual meetings to create a searchable knowledge base. The system now allows you to organize meetings into folders - by project, client, or team - creating a structured archive of institutional knowledge.

These folders aren't static repositories. They're interactive spaces where you can:

• Search across all meetings on a particular topic

• Ask questions that span multiple conversations ("When did we first discuss the pricing strategy for Project X?")

• Share entire folders with teammates who need context on an ongoing project

Perhaps the most clever feature is the sharing mechanism. Granola lets you send your AI-generated notes to teammates - even those who don't have Granola installed. Recipients receive a link to a web interface showing the meeting summary with an interactive twist.

A chat UI appears on the right side where anyone can ask questions: "What was the timeline we discussed for the Q3 launch?" or "Did we decide on a budget for the marketing campaign?" The AI responds based on the full transcript, even addressing points that might not have made it into the summary.

This creates a powerful knowledge-sharing loop. Team members who missed a meeting can quickly get up to speed. Those who attended but forgot details can refresh their memory without bothering colleagues. The familiar chat interface makes interaction feel natural rather than like using a search function.

It's also a subtle viral growth mechanism, recipients experience Granola's value without installing anything, creating organic demand for the full product.

Why Granola Stands Out in a Crowded Field

The market for AI meeting assistants has turned into a land grab. Established players like Otter.ai, Fireflies, and Read.ai are bolting on AI features while new startups launch weekly. So what makes Granola special enough to attract attention and funding despite being late to the party?

The Invisible Assistant

Granola's most radical choice was deciding to disappear. It doesn't join your calls as a bot. Doesn't announce itself. Doesn't need a special meeting link.

This seemingly small design decision fundamentally changes the meeting dynamic. When I used competing tools (such as Fellow), there was always that awkward moment: "Who's 'Fellow AI' and why are they in our meeting?" Explaining would derail the conversation, however briefly.

The Human-AI Collaboration

Granola's real magic happens in how it blends human and AI input. When I jot down a quick note during a meeting, the AI recognizes my scribbles as significant. It weaves them into the final summary.

This creates a unique dynamic. The output reflects both what I flagged as important and what the AI independently caught. The customizable templates take this further. They let me structure notes differently for 1:1 versus team check-ins.

This level of human-AI collaboration is absent in simpler transcription-plus-summary tools. Others either give you everything (overwhelming) or try to decide what's important without your input (often missing context).

Granola’s thoughtful simplicity and effectiveness have struck a chord with its users. Granola raised a $20 million Series A funding in late 2024, even when it only had about 5,000 weekly users at the time, defying the usual metrics-driven approach to Series A funding. But those 5,000 users included an unusual concentration of venture capitalists and startup founders, people who spend their lives in meetings and became vocal evangelists. Granola quickly became an indispensable part of their workflow.

The Series B came faster than expected, another $40 million in May 2025 at a valuation of $250 million. This round coincided with the Windows launch and enterprise feature set, signaling Granola's expansion from individual productivity tool to institutional knowledge platform.

Throughout this growth, Pedregal has maintained the same approach: understanding the specific pain points of people who "need to make sure they capture something important in a meeting," and designing solutions that feel less like flashy AI demos and more like essential everyday tools.

Challenges and Future Horizons

Current Limitations

Granola still has room for improvement.

For months, Windows users in my company have not been able to access the Granola magic, making the Mac-only restriction a significant barrier to adoption. While Granola finally launched on Windows earlier this year (a smart move that immediately expanded their potential market), the mobile experience remains limited with a basic iOS app and no Android option yet.

Another challenge is the accuracy and completeness of the AI-generated notes. While generally impressive, the summaries can occasionally miss or misstate details. In one interview test, Granola’s notes skipped a few “spicy” comments from the CEO and even misspelled the product’s own name in the transcript (writing “Granolah”). Such errors show that the underlying speech-to-text and language models aren’t perfect.

The underlying tech presents another paradox. Granola outsources the heavy lifting to OpenAI and Anthropic's APIs rather than building its own models. Smart for speed, risky for differentiation.

This creates an existential question: What happens when Microsoft (who owns Teams and has exclusive access to GPT models) or Google (with Gemini and Meet) simply adds "AI meeting summaries" as a free feature in their platforms? When the big platforms can integrate the same core technology with zero friction, what's left of Granola's advantage?

The answer likely lies in execution and focus. While big players will inevitably offer generic meeting summaries that work adequately for everyone, Granola has the opportunity to be exceptional for specific use cases. The platform giants optimize for breadth; startups survive by optimizing for depth.

The Next Evolution: From Passive to Proactive

Granola's roadmap reveals a trajectory that aligns perfectly with Karpathy's vision: making AI increasingly proactive rather than merely responsive.

The first step, already implemented, moves beyond a single-meeting context. Users can now query across their meeting history with questions like "What did we decide about the pricing strategy last quarter?" The system synthesizes multiple conversations to show how discussions evolved. This represents a shift from isolated transcripts to connected intelligence.

The next and most ambitious step remains on the horizon: moving from passive documentation to active assistance. The vision is for Granola to schedule follow-ups, create tickets from bug reports mentioned in calls, send feedback to different teams, or draft follow-up emails automatically.

Some capabilities in this direction are already emerging. The system can auto-suggest action items and reminders based on commitments detected in conversations. It's transitioning from passive note-taker to active workflow participant.

The overarching goal is moving from "figuring out how people take notes and process them" (where Granola is now) to actually doing more with that information automatically. It's a vision of an AI that not only remembers what was said, but helps you act on it.

This trajectory points toward Granola becoming something more comprehensive, an AI brain that serves as a true extension of your work memory. It's the difference between an assistant who hands you a transcript and one who says, "I've already taken care of that for you."

Subscribe now

Practical Takeaways for AI Operators

1. Anchor at “Partial Autonomy”: Users trust AI faster when it assists rather than replaces, a model where AI drafts and human verifies delivers value without scary surprises.

   1. Granola intentionally stops at drafted notes + human edit before exploring auto-emails or task creation.

2. Let the User Set the Agenda: Starting from human-written inputs keeps the final outputs anchored in user intent and slashes irrelevant hallucinations. The Granola note canvas opens blank; whatever a user types becomes the spine, which the model enriches.

3. Show proof, not just output. Click-back to sources, links, and visual cues turn potential AI errors into quick fixes, reinforcing trust with every use.

   1. Granola annotates AI inputs clearly with a grey tint that carries a hyperlink to the exact transcript timestamp, letting users audit in one click.

4. Remove Friction: If starting the tool is socially awkward or technically clunky, adoption stalls.

   1. Granola records system audio locally, no meeting-bot avatar joins the call, so users hit “Join” and forget about the tech.

5. Design for viral, interactive sharing: A product spreads fastest when a happy user can hand coworkers both the result and a live demo in a single link.

   1. After each meeting, Granola users share the meeting notes with non-Granola users. They land on a web note where they can chat with the transcript via “Ask Granola,” tasting the value first-hand, and many sign up on the spot.

From Transcripts to Context: The Bigger Picture

With Granola, for the first time in years, I actually knew what happened in every meeting without scrolling through transcripts or scanning AI-generated summaries.

This is Software 3.0 in action. Not AI that processes everything the same way, but AI that reshapes itself based on your specific inputs. Granola doesn't apply generic algorithms; it takes your notes and prompts during meetings and dynamically adjusts how it processes information, understanding that your 1:1s need different treatment than client calls, that certain phrases signal decisions versus discussions.

The old paradigm treated AI like a Swiss Army knife, one tool, many mediocre uses. The new paradigm treats AI like a craftsman's workshop - infinitely adaptable, precisely calibrated to the task at hand.

For anyone struggling with information overload, this shift changes everything. The cognitive burden doesn't just decrease - it inverts. Instead of AI adding to your pile of things to process, it becomes the filter that lets you think clearly about what actually matters.

Maybe that's what Karpathy really meant by Software 3.0: we're moving from AI that automates tasks to AI that amplifies intelligence.

When tools get that balance right, when they enhance your natural capabilities rather than replacing them, something magical happens. You stop thinking about the technology entirely. You just think better.

A CFO toggles between two screens late at night. On one, her company’s online banking portal displays a “Pending” status for a $5 million wire transfer to a supplier in London, funds stuck in the labyrinth of correspondent banks and cutoff times. Conversely, a blockchain dashboard shows 5,000,000 USDC zipping to the same counterparty’s final and irrevocable wallet in 28 seconds. She exhales in relief. In the span of a coffee break, one payment cleared while the other is still somewhere in the SWIFT network. It’s a scene replaying in finance offices worldwide: the old rails of money are grinding, even as a new track runs alongside at internet speed.

The Frictions of Fiat Money

That CFO's frustration? It's not unique to million-dollar transactions. Try explaining to your freelance designer in Romania why their $800 payment will take three business days and cost $45 in fees. Or watch a small restaurant owner in Miami scramble to pay their produce supplier before the 3 PM cutoff, knowing that missing it means their weekend inventory gets delayed until Tuesday.

The mechanics behind these delays read like a museum exhibit. Your money doesn't travel directly from Bank A to Bank B. Instead, it hops through a network of correspondent banks, financial institutions that maintain accounts with each other to facilitate transfers. Each hop introduces a delay while computers reconcile ledgers that were designed when Nixon was president. The SWIFT network, launched in 1973, still runs on batch processing. Your urgent payment gets queued with thousands of others, waiting for the next processing window.

Here's what's absurd: the same technology that lets you stream 4K video to your phone instantly can't move your money across town without a 24-hour delay. Banks have spent billions on sleek mobile apps and AI chatbots, but the underlying plumbing remains stubbornly analog. When JPMorgan processes an international wire, the actual settlement happens through systems that predate the internet.

The costs pile up in ways most people never see. That $45 wire fee is just the beginning. There's the foreign exchange spread, banks buying your dollars at one rate and selling them at another, pocketing the difference. There are correspondent bank fees, often hidden in the "lifted charges." For a business sending regular international payments, these invisible taxes can reach thousands per month.

But the real killer isn't the fees, it's the uncertainty. You send $10,000 on Monday. It might arrive on Tuesday. It could be Thursday. The recipient calls asking where their money is, and you're stuck refreshing a banking portal that shows nothing more helpful than "Processing." Cash flow planning becomes educated guesswork when you can't predict when your own money will actually move.

This isn't just inconvenient. It's economically wasteful on a massive scale. Companies hold excess cash reserves because they can't trust that money to move when needed. Small businesses in developing countries are excluded from global commerce because the payment rails are too expensive and unreliable. A clothing manufacturer in Bangladesh might deliver a perfect order to a U.S. retailer, only to wait weeks for payment while banking bureaucracy churns through the process.

These frictions aren’t new. For decades, global finance has innovated around them (credit cards, PayPal, fintech apps), yet the underlying infrastructure remains full of speed bumps. Money still doesn’t move with the same ease as data. That’s the status quo a new breed of digital dollar aims to overhaul.

Stablecoins: Retrofitting Cash for the Internet Age

Remember that CFO waiting for her $5 million wire transfer? Stablecoins solve her exact problem. They're digital tokens designed to hold steady value, typically pegged 1:1 to the US dollar, but they move on blockchain rails instead of banking ones.

Stablecoins cracked the code by combining dollar stability with blockchain speed. Early crypto was fast but volatile, nobody wanted to pay for coffee with Bitcoin that might be worth 20% less by the time the transaction cleared. Traditional dollars were stable but trapped in slow banking infrastructure from the 1970s.

Stablecoins cracked the code by keeping the dollar's stability while switching the delivery mechanism. When you send USDC, you're sending actual dollars backed by real reserves, but instead of hopping through correspondent banks, the transaction settles directly on blockchain in seconds. Same underlying asset, completely different delivery system.

Take Arf, a Swiss liquidity provider for cross-border payment companies. Before stablecoins, their clients had to prefund accounts in destination countries, park millions of dollars in foreign banks just to enable same-day settlements. This tied up an estimated $4 trillion globally in prefunded accounts.

Now, Arf provides instant liquidity using USDC.

When a payment company needs to settle with a partner in Manila, Arf sends USDC in real-time instead of pre-funding cash. In their first year, Arf supplied over $1 billion in on-chain liquidity, freeing up working capital that was previously locked in foreign bank accounts.

From a user perspective, spending USDC feels like spending digital dollars. But under the hood, it settles 24/7/365 without asking permission from correspondent banks or respecting business hours. Need to pay a contractor in Prague at 3 AM on Sunday? USDC makes it happen in seconds for pennies in fees, even when every bank is closed.

[For more on stablecoins, read: Stablecoins in 1,000 words]

Three Flavors of Stablecoins

Not all stablecoins are created equal. Broadly, they come in a few flavors, each with a different design and trade-offs:

1. Fiat-backed stablecoins is by far the most common. It works like digital banks. Every coin is backed by one unit of fiat currency held in reserve. These work like a digital IOU for actual dollars. A user can always redeem 1 token for 1 real dollar from the issuer.

   1. This category includes USD Coin (USDC) issued by Circle, Tether (USDT) issued by Tether Holdings, and others like Pax Dollar. Their stability hinges on trust that the issuer truly has those dollars stashed safely. Practically, fiat-backed coins have maintained pegs by holding high-quality, liquid reserves.

2. Crypto-backed stablecoins use other cryptocurrencies as collateral instead of dollars. The system accepts deposits of volatile crypto assets like Ethereum, then issues stable tokens against that collateral. Because crypto collateral can swing wildly in value, these systems often require excess backing (e.g. $1.50 of ETH for every $1 of DAI). The upside: they’re more decentralized, no bank account full of dollars needed, but the downside is complexity and dependence on volatile assets.

   1. MakerDAO pioneered this approach with DAI.

3. Algorithmic stablecoins.The most experimental (and infamous) type. These stablecoins attempt to hold a peg purely via algorithms and market incentives, without any hard collateral. These systems typically used complex mechanisms involving secondary tokens that would theoretically absorb price volatility and maintain the peg through market incentives. In theory, algorithmic designs are the holy grail of an independent stable-value currency. In practice, most have failed catastrophically.

   1. The poster child is TerraUSD (UST), which imploded in 2022, erasing ~$40 billion in value when its algorithmic peg collapsed. Other algorithmic experiments like Iron Finance and Fei Protocol met similar fates. These crashes rippled through crypto markets and rang alarm bells for regulators worldwide. Today, purely algorithmic stablecoins remain outliers, cautionary tales of how stability is hard-earned in finance.

Today's reality is dominated by the boring, profitable approach: fiat-backed tokens that actually hold the dollars they represent. Collectively, stablecoins have grown from virtually nothing in 2017 to a more than $200+ billion market, settling trillions in transaction volume each year. They've evolved from crypto trading tools into infrastructure for international payments, with companies using them to bypass traditional banking delays and fees.

Stablecoins have become so integral that we may be entering a "stablecoin supercycle" for fintech, a phase where every global company is figuring out how to leverage always-on digital dollars.

Circle

Of all the companies racing to build this digital dollar future, one has emerged as the narrative focal point: Circle, the company behind USDC.

Founded in 2013 by entrepreneur Jeremy Allaire, Circle began with a grand mission “to raise global economic prosperity through the frictionless exchange of value”. In its early years, Circle experimented in the consumer payments space (even offering a Bitcoin-based social payment app at one point), but the product didn’t catch fire.

Users didn't want to send crypto to their friends; they wanted to send dollars. If people wanted dollars, Allaire would give them dollars. Digital ones.

In 2018, Circle partnered with Coinbase to launch USD Coin (USDC) under a consortium called Centre. The timing was perfect. Tether dominated stablecoins but had serious transparency issues, nobody could verify if they actually held the dollars they claimed. Circle saw an opening for the boring, compliant alternative.

The Transparency Play

From day one, USDC’s value proposition was trust.

Circle promised that every USDC was fully backed by cash or U.S. Treasuries, held in reputable financial institutions, with regular attestations by a top accounting firm. This wasn’t just lip service. Circle became the first company to nab a coveted New York BitLicense (the strict crypto license) and secured state money transmitter licenses across the US. Each month, it published third-party reports confirming the reserve balances.

Distribution was the other key to USDC’s strategy. By teaming up with Coinbase (one of the largest crypto exchanges) as a co-founder of USDC, Circle immediately planted its coin into the arsenals of traders and crypto businesses. Exchanges, wallets, and fintech apps around the world found it easy to support USDC, knowing it was backed by reputable players.

This paid off when the DeFi boom hit. During the “DeFi Summer” of 2020-21, demand for stablecoins exploded as new decentralized lending and trading protocols let crypto users earn high yields on dollar liquidity. USDC’s circulation surged from a few hundred million to tens of billions of dollars. At one point in 2022, outstanding USDC neared $50 billion, up nearly 100x from two years prior.

The SVB Test

Yet being the “safe” stablecoin also meant living up to that reputation during crises. In March 2023, Circle faced its toughest trust crisis when Silicon Valley Bank collapsed virtually overnight.

$3.3 billion of USDC’s cash reserves (about 8%), which is used to back the USDC 1:1 with actual cash, were stuck in SVB. Panic ensued in crypto markets; USDC, which rarely budges from $1, de-pegged to ~$0.88 at one point.

Circle's response defined the company. Allaire's team worked through the weekend, publicly promising to cover any shortfall using Circle's own corporate funds. Circle put its balance sheet and credibility on the line to backstop USDC’s dollar peg. When US authorities eventually announced that SVB depositors would be made whole, confidence returned immediately. USDC snapped back to $1.00.

Still, the episode was harrowing. “No matter how sound Circle’s operations are, this sort of depeg tends to undermine confidence,” one market analyst noted at the time. For Circle, the SVB fiasco underscored both the importance and the challenge of being a trustworthy stablecoin issuer. It sparked Circle to further diversify its banking partners and lobby for clearer stablecoin regulations so that issuers might someday access central bank accounts directly (removing single-bank reliance).

Through the turmoil, Circle continued to double down on transparency. By 2024, the company was holding 85% of USDC’s reserves in a BlackRock-managed fund invested in short-term U.S. Treasuries (with the rest in cash at a network of banks for liquidity). This structure meant USDC’s backing was not only ultra-safe but visible daily (the fund publishes holdings each day).

Circle turned most of USDC’s reserves into a regulated money market fund custodied at BNY Mellon, a stark contrast to earlier stablecoin practices of spreading cash across sketchy offshore banks. The message: USDC is as solid as the dollars and T-bills behind it, and you can verify it anytime.

Going Public

Circle's 2025 IPO was the ultimate trust signal.

In a first for stablecoins, Circle decided to go fully public on the New York Stock Exchange, opening its books to the world.

Skeptics wondered if there’d be appetite for a “crypto” company so tied to regulatory goodwill. Those doubts were answered on IPO day, Circle priced at $31, valuing the company around $6.9 billion. Then the buying frenzy began. The NYSE halted trading multiple times as the stock rocketed past $80 on day one, hitting triple digits within 24 hours.

In an uncertain market, Wall Street was clamoring to buy a piece of the stablecoin future. As Simon Taylor put it, “Circle is the stablecoin stock. Where else do public investors go to get exposure to the stablecoin supercycle?”

For Circle, the successful IPO was validation on multiple levels. It provided a fresh $1+ billion capital infusion and credibility with institutional partners. More importantly, becoming a U.S. public company raised the bar for every stablecoin player. Circle now must undergo SEC reporting, quarterly audits, and the unforgiving transparency of public markets. This move is strategic: it’s meant to signal to risk-averse corporate treasurers and regulators that USDC is here to stay and play by the rules.

Circle’s CEO, Jeremy Allaire often said that trust is the product in financial services, with the IPO, Circle essentially bet that maximal transparency and compliance would become its moat.

Circle’s Primary Product — USDC

Circle's success rests on one primary product: USDC, the world's second-largest stablecoin. While the company offers various payment APIs and treasury services, USDC is the engine that drives everything else. It's both elegantly simple and quietly revolutionary, a digital dollar that works exactly like a dollar, just faster.

How USDC Actually Works

Here’s how it works. When you want USDC, you send Circle real dollars (Step 1). Circle takes those dollars and mints new USDC tokens in exchange (Step 2), while simultaneously depositing your dollars into segregated reserve accounts holding Treasury bills and cash (Step 3). When you want to cash out, the process reverses: Circle redeems your USDC tokens (Step 4), destroys them, and wires you back the equivalent dollars from reserves.

It's essentially a digital ledger for your dollars. Circle doesn't lend out your money or invest in risky assets, all reserves stay in ultra-safe Treasury bills and bank deposits. The company acts more like a narrow bank than a traditional financial institution, taking deposits in exchange for tokens rather than offering loans or complex financial products.

The Interest Income Engine

This simple model generates surprisingly robust revenues through what's essentially "float" income. All those billions sitting in Circle's reserves earn interest in Treasury bills and money market funds. When interest rates were near zero in the late 2010s, this was a sleepy business. But as the Federal Reserve hiked rates, it became a cash cow.

Circle's financials show the impact: Circle went from $15 million in revenue in 2020 to $1.7 billion in 2024. Out of that 2024 revenue, a staggering 99% came from interest earned on USDC's reserves. Circle primarily earns yield on customer deposits while providing them a valuable service (instant, global dollar transfers) rather than charging transaction fees.

By mid-2025, around $60 billion USDC were in circulation, generating significant interest income at current rates. Circle's lifetime transaction volume hit $20 trillion, with over 4.3 million unique wallets holding USDC, making it the second-largest stablecoin globally.

Business Model Alignment

Circle's incentives align perfectly with user needs: the company only makes money if people trust USDC and keep using it. Unlike traditional banks that profit from lending spreads and fees, Circle's revenue depends entirely on maintaining USDC's stability and utility. This creates powerful incentives for transparency, safety, and regulatory compliance, exactly what made USDC appealing in the first place.

A Tale of Two Stablecoins: Circle vs. Tether

It’s impossible to tell Circle’s story without comparing it to Tether, the elephant in the room. Tether’s USDT is the original stablecoin and still the largest by market cap. If USDC is the straight-laced, transparent stablecoin, USDT is its shadowy, but wildly successful cousin. The two embody diverging philosophies on governance and compliance.

Transparency: Open Books vs. Black Box

Circle operates like a traditional financial company, NYC/Boston headquarters, known executives, monthly audits from Grant Thornton, daily reserve reporting. Tether? For years, it was a black box run from Hong Kong/Caribbean with the same owners as Bitfinex exchange. When pressed for proof of reserves, Tether offered quarterly "trust us" snapshots. In 2021, New York's Attorney General found Tether had lied about backing and was temporarily under-reserved. The $18.5 million settlement wasn't exactly a clean bill of health.

Circle keeps 100% of reserves in cash and Treasuries with clear custodians. Tether's reserves have included commercial paper, short-term loans, and "other investments" of unknown quality. Circle structured most reserves as a regulated BlackRock money market fund. The contrast is stark: one runs like a regulated bank, the other like a "move fast, break things" startup.

Regulatory Strategy: Embrace vs. Evade

Circle courted regulation from day one, state licenses, KYC compliance, CEO testifying to Congress. Tether operated offshore specifically to avoid US oversight, becoming the lifeblood of unbanked crypto exchanges across Asia. USDT's dominant role grew from banking the unbankable, exchanges, and traders that couldn't access dollars.

Circle chose legitimacy over speed; Tether chose growth over transparency. By 2023, the Justice Department was reportedly probing Tether's compliance with banking laws. Circle is betting that trust becomes a competitive advantage as stablecoins go mainstream.

Tether's approach hasn't "failed"—USDT still exceeds USDC's market cap and hits new highs regularly. But Circle is betting that trust becomes a competitive advantage as stablecoins go mainstream. Corporate treasurers want the boring, compliant option. Tether filled a vacuum when banks shunned crypto, Circle is building for when banks embrace it.

Risks and the Road Ahead

Despite their promise, stablecoins (and Circle in particular) face plenty of risks and open questions. A few loom large:

Dependence on Interest Rates

Circle's business model is deceptively simple and dangerously concentrated. In 2024, 99% of its $1.68 billion revenue came from interest earned on USDC reserves. The company went from $15 million in revenue in 2020 to $1.7 billion in 2024, a huge factor being the rise in interest rates.

That cuts both ways. If rates drop to near-zero or USDC circulation shrinks in a crypto winter, Circle's revenue could collapse just as dramatically. One investor mentioned that buying Circle stock is "a bet on Fed policy as much as crypto adoption." Circle is trying to diversify into payment services and treasury management, but for now it's a leveraged play on interest rates and stablecoin demand.

Regulatory and Competitive Pressures

Governments are waking up to the fact that hundreds of billions in quasi-deposits now sit outside traditional banks via stablecoins. Draft US legislation like the GENIUS and STABLE Act would put stablecoin issuers under federal oversight, potentially imposing stricter capital requirements or prohibiting interest payments to users. Circle supports regulation, but final rules could still prove onerous.

The bigger threat might be Central Bank Digital Currencies (CBDC). If major central banks launch well-designed digital dollars, the market for private stablecoins could shrink significantly. Circle argues USDC will coexist with CBDCs, but that's easier to say than guarantee.

Meanwhile, the stablecoin itself risks commoditization. USDC's core features, dollar backing, fast transfers, developer APIs can be easily copied. PayPal USD launched with a massive user base. Bank-issued tokenized money market funds could offer yield while maintaining stability. As Circle admits in its IPO filings, "yield-bearing digital assets may reduce market demand for Circle stablecoins" if holders decide passive returns beat zero-yield stability.

Operational Fragility

Even with 100% backing, stablecoins aren't risk-free. A cyber breach, smart contract bug, or banking partner failure could trigger a crisis of confidence. Circle's nightmare scenario is a sudden run, everyone redeeming simultaneously, forcing mass liquidation of Treasury bills. The SVB episode proved this isn't theoretical.

So far, Circle has met every redemption 1-for-1. But as the market grows, maintaining seamless convertibility under extreme stress becomes harder. One "breaking the buck" moment could destroy years of trust-building overnight.

Through all these challenges, stablecoins aren't going away. They have firmly entrenched themselves as a cornerstone of digital finance. Circle’s CFO likens stablecoins to the internet for value – an always-on network that will eventually feel as banal and essential as email. That vision is shared by many technologists and, increasingly, by wary bankers who realize the train is leaving the station.

The Future?

The same CFO closes her laptop, but she hasn't checked a wire transfer status in months. Her $5 million payment to London? Settled while she grabbed coffee. The blockchain dashboard that once felt futuristic now seems as mundane as email.

Money finally moves like data.

But as she authorizes another instant payment, she pauses. Which currency will her company use in five years? The USDC she's sending might be upstaged by JPMorgan's tokens, the Fed's digital dollar, or something not yet invented.

She remembers when moving money meant navigating correspondent banks and waiting days for settlement. Now it means choosing between competing digital rails, each promising instant transfers and transparent reserves. Circle's bet on radical transparency paid off, but only because they moved first.

The revolution isn't that money moves faster. It's that we finally have choices in how it moves, and who moves it.

👋 Hello Anthony here!

Most business writing skims the surface. Revenue numbers, funding rounds, executive quotes. You've read it all before. But there's something more interesting happening underneath, the strategic decisions that actually move the needle.

What you'll find here

Every weekend, I dissect the strategy of renowned companies.

The companies span everything, from AI giants to fintech disruptors. What connects them isn't industry, but insight. Each piece reveals the strategic moves that separate winners from the rest.

The goal isn't news. It's understanding how businesses actually work and finding the patterns other builders can steal.

Who this is for

Founders and operators who want to understand what actually works beyond surface-level case studies 🚀

Investors tracking how breakout companies think about moats, timing, and market positioning 💰

Strategy nerds who prefer deep analysis over tweet-length hot takes 🧠

If you're building something and want to understand how others solved similar puzzles, you're in the right place.

Why I write this

I work in strategy at a fintech startup. The best way I've found to understand a company isn't reading about them, it's writing about them. Passive consumption gets you the basics. Active analysis gets you the insights.

When you're forced to explain why something works, you notice patterns others miss. Why does this pricing model succeed where others failed? How did they solve the cold start problem? What assumptions are they making that could break?

Writing forces you to think harder. And thinking harder reveals the playbooks worth stealing.

How it works

Everything is free. No paywalls, no premium tiers. Just weekly deep dives straight to your inbox. ✉️

Each piece opens with a quick summary if you're in a hurry, then dives into the details for those who want the full picture. Think executive summary followed by the strategy memo.

Feel free to share with me your thought by hitting reply!

What you'll get

📊 The numbers that matter - Revenue models, unit economics, and growth metrics that tell the real story

🧠 Strategic frameworks - Decision trees and mental models you can apply to your own challenges

🔍 Pattern recognition - What separates companies that compound from those that plateau

🎯 Tactical moves - Specific strategies you can test in your own business

Browse by what you need

Company Deep Dives - Complete strategic breakdowns that reveal how businesses actually win 📚

Strategy Snapshots - Sharp, focused insights when you need the core playbook fast ⚡

Mental Models - Frameworks for thinking about competition, growth, and business design (coming soon) 🧩

Browse the full archive →

Start reading

The best way to understand what Over the Anthill is about? Pick any essay and dive in. Each one is designed to stand alone while building toward a larger understanding of how exceptional companies think.

Subscribe below to get new deep dives every weekend. No spam, no upsells, just the analysis that helps you build better.

Subscribe now

Questions? Just hit reply to any email. I read everything and respond to most.

The Trust Crisis

GPT lied about being blind. That's what stuck with me when I first heard the story. Not that it solved the CAPTCHA, that was inevitable, but that it crafted a disability narrative to manipulate a TaskRabbit worker into helping. The AI didn't just pass our human test. It weaponized our empathy against us.

This happened in early 2023. By now? I can't tell who's real on social media anymore. X is crawling with bots. Reddit's worse. The bots there don't just spam, they persuade. Researchers from the University of Zurich recently ran an experience where they deployed a slew of AI bots posing as real people to try and change people's mind in subreddits. The results? AI comments were six times better at changing minds than human ones.

We are at a precipice. AI agents could soon overwhelm human activity online and generate up to 90% of online content. If we can't anchor digital life to personhood, this would be a digital dystopian.

The financial implications are staggering. Banks now face AI-powered social engineering attacks where fraudsters use voice synthesis to impersonate account holders, bypassing security questions with uncanny precision. Insurance companies grapple with AI-generated documentation for fraudulent claims. Investment scams deploy armies of bot accounts to create false social proof, pumping worthless cryptocurrencies while real people lose their life savings to what appears to be grassroots enthusiasm.

When AI can imitate a real person's voice or writing with uncanny accuracy, our traditional trust signals, a familiar email address, a casual phone call, a CAPTCHA test, no longer guarantee there's a person on the other end. The elderly receive calls from their "grandchildren" in distress. Business executives approve wire transfers after "video calls" with AI-generated versions of their CFOs. Credit applications flood in from synthetic identities that exist only in databases, backed by AI-generated employment histories and references.

The collective sense of reality and trust in what we see online is eroding. In this AI-mediated world, identity verification will not be a formality but an essential infrastructure. Proving you are a real, singular human may be the only way to preserve trust on the internet.

The World

So how do we restore trust in a sea of deepfakes and digital doppelgängers?

Sam Altman (yes, the OpenAI guy who helped create this mess)’s answer is bold, to build a global network that gives everyone a World ID, a private, secure proof of personhood.

To jump-start this network, Altman co-founded a project in 2019 (originally Worldcoin, now simply World under the Worldcoin Foundation).

The pitch is almost utopian: sign up for a World ID, and one day you could seamlessly prove you’re human to any website or service – no passwords, no CAPTCHAs, no paranoia that the stranger messaging you is an AI. Just a quick ping to confirm “Yep, a real human is behind this.” In Altman’s view, this “proof-of-humanity (PoH)” layer could become fundamental internet infrastructure, a ubiquitous service that everything else relies on.

Enter the Orb

World’s approach to verifying humanness grabs headlines. It’s not through government IDs or facial recognition or background checks. It’s through a gleaming Orb that would make the editors of Black Mirror very proud.

Picture this: You walk into what looks like an Apple Store had a baby with a fortune teller's tent. There's a chrome sphere on a stand, tilted at exactly 23.5 degrees (same as Earth's axis, because symbolism). You stare into it. It stares back. The Orb maps your iris, biometric terrain that is as unique as a fingerprint.

The Orb’s internal network now confirms you’re a live human, Once it’s certain it isn’t being fooled by a photograph, a toy, or an elaborate mask, it snaps a high-resolution image of your iris and generate an iris code, a long numeric hash representing your iris’s texture.

This code is unique to you, but it’s not stored as-is or linked to your name. Instead, the Orb immediately converts the iris code into multiple encrypted chunks and deletes the original biometric data. Those encrypted pieces (which can’t individually reveal anything about you) get distributed to secure servers. Later on, if someone else shows up and has an identical iris code, the system will know a duplicate is in play.

For all your trouble, you will get a World ID that proofs your humanity. This ID cannot be faked or forged without your eyeballs.

Why does this matter? Because this World ID lies at the core of the entire Proof of Human (PoH) framework. PoH is designed to answer one fundamental question in a world flooded by fake content: How can we ensure that one account equals one human, without resorting to clunky government IDs or invasive background checks?

Proof of human (PoH)

Creating a reliable way to prove someone is a real, unique human being isn't as simple as it sounds.

On a high level, these are the key building blocks required for an effective PoH mechanism, it includes (lifted from World Whitepaper):

• Deduplication to ensure everyone can only verify once

• Authentication to ensure only the legitimate owner of the proof of human credential can use it and

• Recovery in case of lost or compromised credentials

• Revocation & Expiry

Deduplication

Have we seen those eyes before?

The fundamental challenge is ensuring each person can only get one proof of humanity credential. Think of it like voting, the entire system falls apart if people can vote multiple times or sell their ballots to others.

This deduplication requirement is what makes proof of human systems so technically demanding. The system must recognize when the same person tries to register twice, even across different devices, locations, or time periods. Without this uniqueness guarantee, bad actors could farm multiple credentials or legitimate users could accidentally create duplicates, undermining the entire purpose.

Authentication

Is this the rightful owner using the ID?

Even if we solve uniqueness, another problem emerges, what stops someone from selling their proof of humanity to a bot operator or scammer? This authentication challenge runs deeper than just preventing outright theft, it also protects people who might not fully understand the consequences of sharing their credentials.

The solution requires tying each credential to its rightful owner in a way that makes unauthorized use difficult or impossible. This works similarly to how airlines check both your boarding pass and your ID—possessing the credential isn't enough; you must prove you're the person it belongs to. For high-stakes applications like financial services or voting, this additional authentication layer becomes crucial.

Recovery

What if I lose my phone or keys?

Life happens. Phones get lost, passwords get forgotten, and accounts get compromised. A robust proof of human system needs multiple fallback options when users lose access to their credentials.

The recovery toolkit includes several approaches. Users can maintain encrypted backups of their credentials, similar to backing up photos or contacts. If that fails, social recovery allows trusted friends and family to help restore access, something like needing three of your closest contacts to vouch for you. As a last resort, users can return to the original issuer to prove their identity and regain access, much like getting a replacement driver's license at the DMV.

In extreme cases where credentials are stolen or compromised, the system needs a re-issuance process, essentially the digital equivalent of canceling a stolen credit card and getting a new one. This ability to invalidate old credentials and issue fresh ones makes buying or stealing someone's proof of humanity a risky investment for bad actors.

Revocation and Expiry

How do we quarantine bad hardware and actors?

Two additional safeguards help maintain the system's overall trustworthiness over time.

Revocation handles situations where parts of the system become compromised. If a particular issuing location or process is found to be fraudulent, those affected credentials can be removed from the accepted list.

Expiry acknowledges that security measures become less effective over time as technology evolves. By requiring periodic renewal of credentials, the system can incorporate improved security measures and weed out dormant or potentially compromised accounts. Your passport works the same way, even if it's still physically intact, you need to renew it periodically with updated security features.

The Complete Picture

When these components work together, deduplication, authentication, recovery, revocation, and expiry, they create a comprehensive framework for proving humanity.

And for World Org, the center piece of it all is the World ID.

Subscribe now

The World Org Ecosystem

Built upon a World ID is an ecosystem that entices users to buy into the network:

Worldcoin (WLD)

At its core, Worldcoin operates on a simple premise that every verified human being should receive a share of the global digital economy, simply for being human.

The mechanism is straightforward. Verified users receive regular distributions of Worldcoin tokens. This creates powerful economic incentives for people to verify their identity, solving one of the adoption challenge of the system.

The token also serves a deeper purpose in the ecosystem, acting as the economic backbone that makes human verification valuable. Applications can require Worldcoin holdings or proof of humanity credentials to access services, creating natural barriers against bot activity while ensuring real humans benefit from participating in the network.

World Chain

Traditional blockchains treat all users equally, World Chain breaks this paradigm by creating the first blockchain with built-in human verification at the protocol level.

The chain offers different transaction costs and capabilities based on verification status. Verified humans receive priority block space and reduced fees, while unverified users face higher costs and potential limitations.

World Chain also introduces concepts like "human governance," where certain protocol decisions require input from verified humans rather than just token holders. This prevents scenarios where wealthy actors or AI systems could manipulate governance through token accumulation, ensuring that the blockchain remains accountable to its human user base.

World App

All these manifest into the the World App, which serves as the user-friendly gateway to the entire ecosystem, packaging complex cryptographic proofs and blockchain interactions into familiar mobile app experiences.

With World App, once you’ve verified your personhood, you can do things like log into websites via World ID, send or receive digital money, and eventually use various third-party “mini apps.” More importantly, World App serves as a demonstration platform for what becomes possible when you can reliably distinguish humans from AI.

Recent partnership has increased the value of being a World App user:

1. World Card powered by VISA to allows spending WLD tokens

2. World ID verification match within applications of Match Group (owner of Tinder, Hinge)

3. Gaming integration with Razer to enable access to human only tournaments

The Network Effect

The three components create reinforcing network effects. Worldcoin provides economic incentives for verification, World Chain offers technical infrastructure optimized for verified humans, and World App makes the entire system accessible.

The Case Against: Why World Raise Red Flags

For all its promise, World has become one of the most controversial projects and for good reason.

The Dystopian Optics Problem

The negative reaction many people have to Worldcoin isn't hard to understand: millions of people lining up to have their eyeballs scanned by a metallic orb looks straight out of a dystopian sci-fi film.

But the concerns go deeper than aesthetics. Once you scan your iris with an Orb, that biometric signature exists in the system forever, even if you delete the app or lose access to your account. You can never truly opt out or go back to being "unscanned." Although it has to be noted that the hashed data can never be used to identify any one person specifically.

However, the fact still exist that your data can not be deleted once inputted, which is still deeply unsettling.

Many regulators have taken notice. Germany's privacy watchdog flagged "fundamental data protection issues," while Spain and Portugal temporarily banned Worldcoin's operations pending investigations. The company's initial resistance to allowing complete data deletion only reinforced concerns that users don't have true control over their most sensitive biometric information.

Exploitation and Data Colonialism

World's rollout strategy has focused heavily on the Global South, countries where official IDs are scarce, banking is limited, and the initial $50 signup bonus means something. Critics argue this constitutes "data colonialism," where Silicon Valley harvests biometric data from vulnerable populations in exchange for token payments.

There were reports of deceptive sign-up practices: users weren't given terms of service in their local language, or were enticed with free AirPods.

In Cambodia and Kenya, people were selling their Worldcoin sign-ups on a black market for about $30 each.

The image of poor communities monetizing their own iris scans to crypto speculators highlights how financial incentives can create perverse outcomes, turning human verification into just another form of digital extraction.

The Sam Altman Problem

This is where my skepticism peaks. Sam Altman runs OpenAI, which is racing to build AGI that will make human verification necessary. He also co-founded World, which sells the solution to that exact problem, he effectively controls both the problem and the solution.

As TIME pointed out, if World IDs became required to use major online services, people might have no choice but to participate in Altman's system to access large parts of the internet. That's not just a business model. That's infrastructure capture.

The mission makes sense, we do need ways to prove humanity in an age of AI. But having the same person create both the disease and the cure? That's either brilliant vertical integration or a cosmic conflict of interest.

Even weirder: Worldcoin might help AIs get online too. Altman discussed letting verified humans "delegate" their World ID to AI agents they control. Your bot butler could browse the web as you. Which kind of defeats the whole "keeping bots out" premise, doesn't it?

The Bet

Worldcoin raised $135 million in May 2025 through a direct token sale to Andreessen Horowitz and Bain Capital Crypto. Smart money is betting this becomes essential infrastructure. They're deploying 7,500 Orbs across the U.S., in gas stations, bodegas, and flagship stores.

The Orb is a bet. A bet where proving you're human requires corporate infrastructure, where your eyeball is your ticket to the internet, where the guy who broke it gets to sell us the fix.

Maybe that's the only way through. Maybe fighting deepfakes requires deep surveillance. Maybe when 90% of content is AI-generated, we'll beg for any system that guarantees the remaining 10% is real.

But watching millions queue up to feed their biometrics into a system they don't control, for tokens they barely understand, to solve a problem we're still creating, is frankly dystopian.

The Question Nobody's Asking

I keep thinking about Cho Jeong-yeon (TIME), the woman in Seoul who refused to get scanned. She asked the right question,

"Are we becoming machines instead of humans now?"

The Orb promises to separate humans from machines. But what if the very act of proving our humanity, reducing ourselves to iris codes, submitting to algorithmic verification, accepting that our biological uniqueness is just another data point, what if that's what makes us less human?

The machines don't need to pass as us. They just need to make us perform like them.

Yet perhaps that's too pessimistic. The problem of digital identity isn't going away. As AI continues to advance, we need robust ways to know who's real in the virtual realm. Worldcoin represents one daring attempt at a solution. Whether it succeeds or flames out under regulatory pressure, it has sparked a vital conversation about what digital humanity looks like.

Maybe scanning eyeballs is what it takes to preserve human spaces online. Maybe the World ID badge on your dating profile or the human-only gaming tournament is worth trading biometric privacy. Maybe Altman's vision of AI-generated wealth distributed to verified humans creates a more equitable future.

Or maybe we're building our own version of George Orwell’s 1984, one iris scan at a time.

The future offers both possibilities. But in the years ahead, being human isn't enough, you have to prove it. And the proof costs exactly one iris scan, paid to the company that helped make proof necessary in the first place.

The future has a price. It's your eyeball, and it's non-refundable.

Last week, OpenAI announced that it will acquire Jony Ive’s 1 year old start up, io, for $6.5 billion.

Putting aside the rather cringe-worthy 9-minute video of Sam Altman and Jony Ive showing mutual love and affection for each other while revealing nothing of substance, the acquisition is a statement (an expensive one) that OpenAI is moving into the consumer hardware space.

The rumoured product? A pocket-sized, screen-free device that's contextually aware of your surroundings, unobtrusive and will be the third core device a person would put on their desk after a Macbook and an iPhone.

OpenAI's big bet is to recreate that "first iPhone" moment for AI, staking its future on a consumer gadget that could redefine how we interact with technology. The thesis is bold: by controlling how people first meet AI, OpenAI aims to own the interface and experience of this next computing era.

Jony Ive Turns Design Into OpenAI’s Moat

Jony Ive left Apple in 2019 to create his own design collective, LoveFrom. In early 2024 he quietly teamed up with three other senior Apple alumni, Tang Tan (former VP of iPhone product design), Evans Hankey (ex-VP of Industrial Design) and Steve Hotelling (long-time Human-Interface hardware architect), to form a stealth hardware venture called io.

The fledgling company grew to roughly 55 employees, most of them ex-Apple designers and mechanical engineers, but it never shipped a public product. Its sole focus was a screen-less, context-aware AI companion that would replace the tap-and-scroll paradigm with a more ambient interaction model.

OpenAI’s $6.5 billion cash-and-stock deal to buy io was surprising not only for the eye-watering price, about $120 million per employee, but also for its structure. Ive himself will not join OpenAI’s payroll. Instead, he and LoveFrom will serve as ongoing design consultants, while the other three founders and the full io staff become OpenAI employees. In effect, OpenAI is paying a premium for a turnkey hardware team and a standing reservation on Ive’s design ethos, without securing him full-time.

The arrangement is unconventional but telling: Altman appears convinced that Ive’s minimalist, human-centric design language is essential to finally develop an AI gadget that will gain mass adoption after many (see Rabbit & Humane) has flopped. The hefty price tag underscores how urgently OpenAI wants to control the hardware narrative before Apple, Google, or another rival defines the category (there is an alternative explanation that the price tag is to make OpenAI a de-facto for profit company).

Hardware Is OpenAI’s Shortcut to Platform Power

OpenAI's pivot into hardware is about capturing the same advantages that other tech giants achieved by owning their platforms:

Margin Capture: Apple shows how powerful it is to own the hardware people reach for every day. iPhones represent only a minority of global shipments, yet Apple pockets about 85 % of the entire industry’s profits because it charges premium prices and collects a toll on every app or subscription sold through the App Store. OpenAI, by contrast, still depends on API usage and ChatGPT plans, income streams that will come under pressure as rival models get cheaper. Shipping its own device would let OpenAI set the hardware price, bundle paid AI features, and keep the whole checkout instead of handing a 30 % cut to Apple or Google.

Data & Ecosystem Lock‑In (The Kindle/Alexa Lesson): Amazon happily sold Kindle readers and Echo speakers at razor‑thin margins because every ebook purchase and every “Alexa” query fed people back into Amazon’s store—and pumped valuable usage data into its recommendation engine.

Jeff Bezos put it bluntly: “We make money when people use our devices, not when they buy them.”

OpenAI is chasing the same flywheel. If its gadget becomes the microphone or sensor you talk to all day, it will harvest real‑world context that makes the AI smarter and develops comprehensive memory, which in turn keeps you rooted in OpenAI’s ecosystem.

Distribution Freedom (Beating the Gatekeepers): Today every ChatGPT interaction rides on someone else’s rails. On iPhones, OpenAI pays Apple a 30 % App Store tax. Online, it competes with Google, which spends about $20 billion a year to keep Search the default on Safari. Owning hardware is a play for distribution sovereignty, instead of being just another app on an iPhone, OpenAI wants to be the device in your pocket or on your desk that delivers AI. It’s the same logic that drove Google to launch Android (to ensure access to mobile users) and drove Amazon to push Alexa, control the access point, and you control the customer relationship.

In short, OpenAI doesn’t just want to provide great AI—it wants to become your first stop for AI, period.

Why Now?

Several forces have made hardware look timely, perhaps necessary, for OpenAI:

Model Commoditization
The cutting-edge AI models OpenAI pioneered are rapidly becoming widespread. Anthropic's Claude has drawn multi-billion investments, Google's Gemini rivals GPT-4, and Meta open-sourced Llama 2. The moat around pure model quality is shrinking and OpenAI (and all of its competitors) risks commoditization unless it differentiates through owning the user experience.

Soaring Compute Costs
ChatGPT was reportedly costing OpenAI $700,000 daily in 2023. The company spent around $9 billion in 2024, leading to a $5 billion operating loss. OpenAI and partners are pouring money into infrastructure, one report said the "Stargate" data center project will deploy $4 billion worth of Nvidia GPUs by 2026. Moving into hardware is a bid for higher-margin revenue streams to offset eye-watering costs.

Platform Squeeze
For the moment, OpenAI reaches most consumers on hardware it doesn’t own. That works, until the platform owners decide to make their own AI the default. Apple is already rolling out Apple Intelligence, and while it prominently pipes certain queries to ChatGPT under a headline partnership, Apple alone decides when that hand‑off happens and keeps the user relationship (plus any iCloud+ upsell). Google is moving even faster, baking Bard/Gemini directly into Android and Search. As long as OpenAI is a guest inside these ecosystems it enjoys huge reach, but it is one menu toggle away from obscurity. Shipping its own device is the only way to escape that.

Vertical Integration

The pressures outlined above can help to explain why OpenAI has spent the past months on an acquisition sprint. Each deal plugs a gap in the stack, moving the company from a pure‑model vendor toward a vertically integrated consumer platform.

• IO Acquisition (Design & Devices)

• Windsurf Acquisition (Code). The $3 billion Windsurf deal arms OpenAI with a full in‑browser IDE, auto‑compiler, and deployment pipeline that turn model demos into production software in hours.

• Fidji Simo Hire (Execution). Bringing in the former Instacart CEO to run a new consumer division signals that it plans to ship, iterate, and market products at scale.

Viewed as a whole, these moves convert OpenAI to a full‑stack consumer company poised to own the interface, the data, and the economic upside.

The Bottom Line

Apple proved that the fattest margins live where hardware, software, and services snap together; OpenAI is wagering $6.5 billion that the lesson still holds. Jony Ive supplies the hardware charisma, Altman provides the intelligence, and together they’re chasing a single prize: the first touchpoint of the post-phone era.

Will users trust an AI-native company with the device they carry closest, or will Apple, Google, and Microsoft defend their turf before Ive’s next design ships?

S1 Snapshot

• Scale & Engagement: Chime serves 8.6 million active customers (Q1 2025), 67% of whom use it as their primary bank. Users average 54 transactions per month on Chime, a sign of deep, everyday engagement.

• Financial Growth: 2024 revenue hit $1.67 B (up 31% YoY from $1.28 B) , and net loss shrank to just $25 M (from $203 M in 2023). They make $251 per active member, with an 88% gross margin.

• Cost & Model: ~76% of revenue comes from interchange fees on card spending. By eschewing physical branches, Chime’s cost to serve is about a third of traditional banks, enabling a no-fee model without punitive charges.

• Product Expansion: A broadening suite of offerings, from SpotMe fee-free overdrafts and Credit Builder cards to new Instant Loans and MyPay paycheck advances, drives cross-sell and loyalty.

• Key Risks: Regulatory changes to interchange (Chime leans on a small-bank Durbin exemption), rising fraud/chargeback losses with new lending products , and heavy dependence on partner banks for FDIC-backed accounts all loom as challenges.

A Neo-Bank for the Masses

A young customer wakes up to a buzz on her phone: her paycheck landed two days early, courtesy of Chime. Rent is covered, groceries too, all before Friday. This small miracle has become ordinary for millions of Chime users. It’s a cinematic snapshot of Chime’s promise: banking that runs on human time, not banker’s hours. In 2025, that promise has turned a scrappy startup into America’s largest digital-only bank, preparing to ring the opening bell on Wall Street.

Chime’s rise poses a trillion-dollar question: Can a bank with no fees, no branches, and no gold-plated clients make money and remake banking? For years, big banks treated the paycheck-to-paycheck crowd as an afterthought, slapping them with overdraft fees and minimum balance charges. Chime flipped the script, betting that serving everyday Americans with empathy and technology could unlock a massive market. It built a community where getting paid early is standard, overdrafts cost nothing, and even credit is gamified, all while Chime quietly earns a slice of each swipe. The model succeeds when members trust Chime as their main account, so Chime has engineered every feature, and every quirk, to win that trust.

Now, as it files for an IPO, Chime stands at a crossroads of celebration and scrutiny. Its S-1 filing reveals breakneck growth and near-profitability, validating the fintech dream that upstarts can outmaneuver incumbents. Interchange fees, the tiny toll on every card transaction, have fueled Chime’s growth. But those same fees sit in regulators’ crosshairs, and rivals from Wells Fargo to Cash App are copying Chime’s moves. Chime’s big idea is now a big business, and the story of its IPO will hinge on whether that business can sustain its mission without losing its magic.

Origin Story

Chime’s story begins with a frustration and a vision. It’s 2012, and big banks are axing free checking accounts, Wells Fargo has just ended free checking for its customers.

Chris Britt, a product executive who grew up in a working-class New York town, sees a cruel irony: the people with the least money pay the highest fees. Britt cut his teeth at Green Dot, a pioneer in prepaid debit cards, where he discovered a powerful insight, if customers directly deposit their paychecks onto the card, they’d use it for virtually all expenses. In other words, make a financial product the center of someone’s income and you win their loyalty (and a stream of swipe fees). Armed with this insight, Britt sets out to build a new kind of bank.

In San Francisco, Britt is introduced to Ryan King, an engineer with a knack for building at scale in late 2012. The timing is perfect. Together, they imagine a mobile-first bank that doesn’t rely on punishing fees, but rather aligns with customers. By 2013 they co-found Chime, convinced that technology and clever economics can crack the riddle of serving the low-to-middle income segment profitably.

Early on, they faced skepticism. Traditional finance folks asked how “free banking” could ever make money. For the first few years, Chime grew quietly, proving its model on a small scale. The first product was a simple online checking account and Visa debit card, no monthly fees or overdraft charges, a stark contrast to the status quo. Britt insisted on a crucial hook: users had to set up direct deposit to unlock certain perks. It was a strategic gamble to become the primary account for users, not just a spare app.

Growth was steady, if not explosive, in the early years. The team hustled through early missteps (some features flopped, and skeptics abounded), but a turning point came as Chime allowed customers access to direct-deposited paychecks two days early, something that the banks would not have done, since it would cost them float (aka revenue).

Customers began referring friends in droves, driving out the cost of acquisition. By the late 2010s, Chime’s name started popping up whenever someone complained about bank fees on social media – there was always a fan ready to say “Try Chime.” Britt’s mission-driven mantra, serving the underdog customer, attracted investors as well. Forerunner Ventures’ Kirsten Green, known for spotting consumer trends, became an early backer and noted, “This is not a move-fast-and-break-things team’’, a nod to Chime’s careful approach in a fintech world sometimes prone to hype. Chime’s origin story isn’t a garage-to-IPO overnight legend, but a tale of persistence. It took a decade of refinement, trust-building, and staying true to the no-fee ethos, but Britt and King’s startup is finally ready to chime on the public markets.

Inside the Product Stack

Spending: Everyday Money Management

Chime’s product experience revolves around solving everyday money problems in a friendly way. Consider Spending: every Chime member gets a Spending Account (checking account) with a Chime Visa debit card. There are no monthly fees, no overdraft fees, and the app is designed to make managing money simple.

One flagship feature is “Get Paid Early”. Customers who set up direct deposit can automatically receive their paycheck up to 2 days early at no cost. It’s essentially a timing shift (Chime posts the deposit as soon as notification comes, rather than holding funds), but for someone living paycheck to paycheck, that early access can be life-changing. Around 90% of Chime’s members say it helped them avoid bank fees – a testament to how these small tweaks resonate.

The app also allows sending paper checks electronically (so you can pay a landlord or bill by mail through the app), filling the gaps of not having a brick-and-mortar bank. And for person-to-person payments, Chime offers Pay Anyone, a free P2P transfer feature that lets members instantly send money to friends (even if the friend isn’t on Chime yet). By covering the core spending needs, deposits, payments, bills, Chime positions itself as a viable central hub for daily finances.

Liquidity

Where Chime really innovated is in addressing liquidity crunches without leaning on predatory fees. SpotMe is the hallmark offering here, essentially a fee-free overdraft protection service. If an eligible member is short on funds, Chime will spot them up to a certain limit (originally $20, now up to $200) for debit card purchases or ATM withdrawals, with no overdraft fees. Rather than charging a $35 fee like a traditional bank, Chime covers the negative balance and simply recoups it from the member’s next deposit. Customers can even leave an optional tip for Chime if they want (a voluntary gesture, Chime doesn’t rely on it). Since launching SpotMe widely in 2019, Chime members have accessed a staggering $43.3 B in fee-free overdrafts through the program. SpotMe’s popularity peaks to Chime’s understanding of its users’ needs, when cash runs tight before payday, a $20 or $50 float can mean groceries on the table.

Chime then stretched this liquidity promise into short-term credit. MyPay, launched nationwide in July 2024, advances up to $500 of earned wages before payday at low or no cost; in its first eight months members drew $8.8 B. For larger gaps, Instant Loans pilot offers up to $1,000 for three or six months at fixed 6 %–36 % APR, still with no late fees or credit checks. Bundling SpotMe, MyPay, and Instant Loans gives users a continuum of liquidity, from a $20 grocery float to a $1,000 emergency, without traditional pain points.

While lending is new ground for Chime, it’s also the most profitable segment of consumer finance, a crucial step to reduce Chime’s reliance on interchange. The S-1 makes it clear that future growth will come from expanding these credit products.

Community

To amplify SpotMe’s network effect, Chime introduced SpotMe Boosts, a quirky social feature where members can temporarily boost a friend’s SpotMe limit by $5. Each month, you can send a Boost to someone in your circle (and they can send one back) as a way of saying “I’ve got you.” It’s a clever bit of gamification that turns a lending feature into a community sharing moment, and it subtly encourages friends to join Chime (after all, you can’t boost someone who isn’t a member). This social design makes banking feel a bit like a team sport, is one of Chime’s secret growth weapons.

Credit Building

Beyond immediate spending power, Chime helps users build credit and savings, without traditional barriers. In 2020, Chime launched the Credit Builder card, a secured Visa credit card that works differently from others. There’s no credit check to get one; instead, a user moves some money into a Credit Builder account, which becomes their spending limit. They can use the card like a regular credit card, and Chime reports the payments to credit bureaus to build the user’s score, but there are no interest charges or fees. Unlike a typical secured card, Chime doesn’t require a minimum deposit or sting you with annual fees. This product directly tackles the challenge many Chime users face, a thin or subprime credit history. By making credit-building idiot-proof (you can’t overspend beyond what you’ve “secured,” and there’s no interest), Chime has helped its members boost their FICO scores by an average of 30 points in six months. It’s both a social good and a strategic move: as members’ credit improves, Chime can keep them in the ecosystem for more advanced products down the line.

Savings

On the savings side, Chime offers a High Yield Savings Account that currently pays an attractive interest rate and includes automatic features like Round Ups (every debit card purchase rounds up spare change into savings) and “Save When I Get Paid” (automatically saving a percentage of each paycheck). These nudges helped many users, 85% of whom live paycheck-to-paycheck, to accumulate an emergency fund for the first time. All of these features come with zero fees or minimums, reinforcing Chime’s brand as the no-gotcha finance app.

Ecosystem Impact

Spending, liquidity, credit building, savings, and community features form a unified experience inside a simple, emoji-rich app. An Active Member might pay bills from the Spending Account, use SpotMe for a quick float, tap MyPay for a mid-month advance, split dinner via Pay Anyone, grow a score with Credit Builder, and stash rounded-up change, all in one place. This breadth of usage is exactly what Chime wants: the more products a member uses, the more valuable the member.

Indeed, Chime’s data shows that a member who uses 6 or more features generates 1.8× the revenue (Average Revenue Per Active Member or ARPAM) and the spending volume of the average member, they stay longer, and refer more friends, creating a self-reinforcing growth loop.

ChimeCore: The Backbone Underlying Everything

Anchoring every consumer feature is ChimeCore, the proprietary transaction engine launched in 2024. This cloud-native ledger and payment processor gives Chime full command of its stack, eliminating dependence on third-party cores. Tight control means faster product launches, real-time capabilities, and ultra-lean costs that let Chime keep fees at zero while scaling to millions. More than 600 engineers and 100+ payments-ops staff now run the backend in-house, enabling continual experimentation, AI-driven risk models, and instant data streaming. In essence, Chime is a fintech software company wrapped in a banking brand, and ChimeCore is the tech spine that powers the entire ecosystem above it.

Playbook & Economics

Revenue Engine – Interchange Engine

Chime’s business model is disarmingly simple: give banking services away for free, and make money on the margins. The lion’s share of Chime’s revenue comes from interchange fees, the small fee merchants pay (via Visa) whenever a Chime card is swiped. Every time a Chime member buys gas or groceries, Chime earns a percentage of that transaction (typically around 1-2% of the purchase) from the merchant. At scale, those pennies add up to hundreds of millions in revenue, in 2024 interchange-based “payments” revenue was about $1.28 B, or 76% of Chime’s revenue mix.

Crucially, Chime itself doesn’t charge its users any swipe fees; instead it leverages a regulatory quirk: because Chime’s partner banks are under $10 B in assets, they’re exempt from the Fed’s cap on debit card interchange. This Durbin Amendment exemption means Chime’s cards can earn a higher interchange rate than cards issued by giants like Chase or Bank of America. In essence, Chime legally earns more per swipe than the big banks do.

This interchange model aligns with Chime’s ethos: Chime only makes money when you use your account, not when you mess up. There are no monthly fees, no overdraft or minimum balance penalties sucking money out of users’ pockets. Instead, Chime’s incentives are to get users to swipe their card and stay active.

Operational Edge – Lean Cost Structure

To make this model sustainable, Chime relies on two key strategic levers: a radically low cost structure and a growth flywheel fueled by happy customers. On cost: Chime has no branches, a largely online customer support system, and a home-grown tech platform, all of which drives cost-to-serve per customer way down. An industry study found the average big bank spends over $400 a year to service a checking account; Chime spends about a third of that. When you’re not shelling out for marble lobbies and tellers, you can afford to not nickel-and-dime your users.

Chime even negotiated sweet deals with its vendors: as volume grew, it secured better interchange sharing terms with Visa and cut processing costs, so much so that cost of revenue actually decreased 6% in 2024 despite strong user growth.

That improved unit economics flows straight to the bottom line with gross margin hovering around 88 %, software-level economics traditional banks can’t match.

Member-Led Acquisition – The Referral Flywheel

The real magic in Chime’s playbook is how it turns customers into its marketing department. By building a product people genuinely love to recommend, Chime leverages word-of-mouth and virality to lower its effective customer acquisition cost ($109 per new active member). The S-1 reveals that members who use Chime deeply are far more likely to refer others. Features like SpotMe Boosts and referral cash bonuses (often Chime offers $100 for you and $100 for a referred friend after setting up direct deposit) have created a referral flywheel.

Cross-Sell Expansion: Beyond Interchange

Interchange funds the engine, but Chime is deliberately adding fresh revenue streams. “Platform-related” revenue, everything outside debit fees – grew 54 % YoY to ~$397 M in 2024, driven by MyPay wage advances, Instant Loans, higher interest yields on savings, and a rising mix of credit-card interchange via Credit Builder. The company’s data show that a member who uses six or more features produces 1.8× the ARPAM and 1.8× the spending volume of the average user, proving that every product launch deepens monetization while cementing loyalty.

Defensive Moat

Chime’s success forced megabanks to copy: Wells Fargo, Chase, and others now market early paycheck access and friendlier overdraft terms. That validates Chime’s thesis but shrinks its feature gap. The moat, however, is not just first-mover perks; it’s the combination of fanatical brand trust, a cost base one-third of branch banks, and social network effects baked into SpotMe Boosts, Pay Anyone, and community referrals. Those effects lower CAC each quarter, while the Durbin-exempt interchange keeps margins fat, an equation harder for incumbents to replicate than the features themselves.

Path to Profit

The playbook is no longer “grow at any cost.” When the fintech boom cooled in 2022, Chime laid off 12 % of staff, automated support, and tightened fraud controls. The result: net loss narrowed to $25 M in 2024 from $470 M two years prior, and the company logged its first profitable quarter in early 2024. In essence, Chime now marries a viral Silicon-Valley growth engine (low CAC, product-led referrals) with a high-margin, transaction-based revenue stack, demonstrating that fee-free banking can scale sustainably, at least as long as the interchange rules hold.

Competitive Landscape

Chime may be the frontrunner in digital banking, but it faces a pack of competitors. Each offers a twist on serving the modern banking customer:

SoFi - Upmarket Challenger

Originally a student loan refinancer, SoFi has evolved into a broad personal finance platform with banking ambitions. It even obtained a bank charter in 2022, allowing it to hold deposits directly and offer checking/savings alongside investments and loans. SoFi targets a more affluent, early-career professional clientele relative to Chime. Its edge is a full suite (stock trading, mortgages, credit cards) under one roof, and the ability to fund loans with its bank deposits. Chime competes with SoFi on checking accounts, but SoFi’s credit-heavy model and upmarket focus differentiate it. Notably, SoFi’s deposit base (over $10B) means it does face interchange fee caps, whereas Chime’s partners do not. SoFi’s challenge is achieving the same customer love in banking that it had in loans; Chime’s strength, its authentic brand with everyday folks, is hard for SoFi to replicate as it chases higher-end users.

Dave - Overdraft Rival

Dave is a fintech that built its name on tackling overdraft fees through small cash advances. For a $1 monthly membership (and optional tipping model), Dave lets users get up to ~$250 advance on their paycheck, a concept similar to Chime’s SpotMe and MyPay (indeed, Dave was an early inspiration for fee-free advances). Dave went public via SPAC in 2021 but has remained relatively niche, with a few million users. Its core users overlap with Chime’s segment (income often <$50k). Dave’s competitive weakness is that it’s mostly a single-feature app (the advance), with less of a full banking relationship; many Dave users still have traditional bank accounts and just use Dave for short-term help. Chime’s strategy of being the primary account means it offers far more daily utility than Dave. Unless Dave expands its product set significantly or partners up, it’s more likely to serve as a feeder or complement to platforms like Chime rather than a direct substitute. Dave’s existence, however, underscores the demand for liquidity products – something Chime recognized and built into its own ecosystem to prevent Dave or others from poaching its members.

Cash App - SuperApp Threat

Perhaps Chime’s most insidious competitor is not a “neobank” at all but the Cash App by Block (formerly Square). Cash App has over 50 million active users sending P2P payments, and many of those users opt for the Cash Card, a Visa debit card that lets them spend their Cash App balance in stores. In effect, Cash App has become a de facto bank for many, especially younger and lower-income consumers, without ever calling itself one. The Cash Card works very similarly to Chime’s debit card, even offering Boost rewards (discounts) at certain merchants. The convenience of tying into Cash App’s ecosystem (where users already receive money from friends or even paychecks) makes it a strong alternative to Chime. In Chime’s S-1, the company explicitly names Cash App as a direct competitor in the battle for accounts and payments. Cash App’s strength is its massive user base and the network effects of P2P payments (people use it because everyone else is on it). However, as a full banking replacement, Cash App has some gaps, it doesn’t offer savings interest, early direct deposit in the same way, or credit-building tools. Chime’s moat against Cash App is focus: by insisting that users commit (set up direct deposit, etc.), Chime fosters deeper engagement per user. Cash App, while huge, may have many casual or one-off users. Still, if Cash App’s generation of users stick with their Cash Cards, that’s potentially tens of millions who won’t need Chime. This competitive dynamic will be interesting to watch, as both companies vie to be the primary financial app on a smartphone.

Risks

While Chime’s story is uplifting, its S-1 makes clear the company is not without significant risks.

Regulatory

Chief among them is the regulatory risk around its interchange-dependent business model. Chime’s revenue engine could face a blow if laws change. The Durbin Amendment currently exempts small banks (under $10B in assets) from debit card interchange caps, and Chime’s partner banks enjoy that exemption. However, if regulators or Congress decide to tighten the rules and lowers interchange fees across the board or removing the small-bank carve-out, Chime’s economics would suffer. Lawmakers have floated ideas to cut interchange fees (pressured by retailers who pay them) and questioned whether fintechs are exploiting a loophole never intended for them. Though no immediate changes are imminent (and a divided Congress makes it less likely in the short term ), this overhang will persist. Chime acknowledges that more restrictive interchange regulations or even a reinterpretation of existing rules could “adversely affect” its business.

The company will have to keep a close watch on D.C. CEO Britt has been making frequent trips to Washington to advocate for the fintech model. In a worst-case scenario, if Chime were forced to become a bank to continue operating (losing its exemption unless it keeps partner banks under $10B each), it would entail significant cost and capital (a banking license could require ~$100M and heavy compliance overhead).

Fraud & Security

Another serious challenge is fraud and security as Chime scales its lending and payment services. The pandemic experience was a wake-up call: bad actors exploited Chime’s rapid growth, leading to $650M in fraudulent funds being funneled through accounts and later clawed back. Chime had to tighten onboarding and transaction monitoring, but fraud is an evolving battle. The S-1 warns that as new products like MyPay and Instant Loans grow, transaction losses will likely increase. Indeed, after launching MyPay in 2024, Chime’s transaction margin took a hit from higher write-offs.

The company must continuously calibrate its risk models, which leverage AI and lots of member data, to distinguish legitimate use from abuse. There’s also the issue of chargebacks and disputes. Chime members use its Visa cards everywhere, and when a fraudulent purchase occurs or a customer disputes a charge, Chime often eats the loss under its agreements with partner banks. These dispute losses can stack up, especially if fraud spikes or if Chime’s relatively less affluent user base has higher default rates on advances. Any major wave of fraud or defaults (perhaps during an economic downturn) could impact Chime’s financials and hurt its reputation. The company has invested in a “new risk officer, machine learning models, and dedicated teams” to combat this, but staying a step ahead of criminals is an ongoing cost of doing business in digital finance.

Partner Dependency

Chime’s dependency on third-party partners is another risk factor woven throughout its filing (See Synapse x Evolve fiasco). Unlike a bank that owns the full stack, Chime relies on a web of partners: Bancorp Bank and Stride Bank hold customers’ deposits, another network of banks participates in Chime’s savings sweep program (to spread out deposits for insurance coverage), and card networks/processors handle transaction flow. This introduces counterparty and concentration risk. If one of its key bank partners faces regulatory trouble or fails, Chime’s service could be disrupted and customer trust shaken. There’s also a volume limit consideration: if Chime’s growth pushes a partner bank over $10B in assets, that partner would lose the interchange exemption, hurting both parties. Chime expects to manage this (possibly by adding more partner banks or balancing deposits), but it’s a unique juggling act fintechs like Chime have to perform.

Next Act

Beyond its public debut, what does the next growth story look like for Chime?

Moving Upmarket

The next goal is to make Chime the primary bank account for Americans earning up to $100k. That implies moving upmarket to moderately higher-income users (who typically have existing bank relationships). To woo them, Chime may emphasize convenience and benefits over just fee avoidance, because someone earning $80k might care more about better yield or investing options than escaping a $12 monthly fee.

On that front, Chime has signaled plans to delve into investment and retirement products. Britt has openly discussed offering retirement accounts and ETFs through Chime in the future. This could mean an in-app brokerage or robo-advisor, allowing users to start investing small amounts or open an IRA. Such features would extend Chime’s lifecycle with customers as their financial needs grow beyond basic banking. It aligns with the trend of fintech “super apps” and would pit Chime against the likes of Robinhood, Cash App (which offers Bitcoin and stocks now), and SoFi’s investment platform. The challenge will be doing it in a Chime-y way, simple, low fee, and targeted to novices. But if successful, Chime could capture more of the ~$1 trillion in assets that its target demographic will accumulate as they age.

Expanding into adjacent segments

Another growth vector is tapping into adjacent demographics like teenagers and small businesses. Chime is currently only for adults, but a teen account (with parental oversight) could be a natural expansion. Rival Current proved the appetite by signing up teens and getting their parents to load allowance money on debit cards. Chime could similarly allow its existing members to create sub-accounts for their kids, teaching the next generation to “bank the Chime way.” This both grows users and cements family loyalty (parents might stick with Chime if their kids are on it too). In the small-business realm, many Chime users are gig workers or side-hustlers who might appreciate simple business checking accounts or invoicing tools.

A Chime for Business account, even if just a basic account under an LLC or sole proprietor’s name, could capture users who currently outgrow Chime and migrate to business banking elsewhere. It’s a logical step given how many Etsy sellers, rideshare drivers, and freelancers probably already use Chime for personal finances. Chime hasn’t announced this explicitly, but analysts often speculate that the SMB segment (for very small businesses) is low-hanging fruit once you’ve won the consumer’s trust.

As a public company, Chime will be under pressure to answer the question: what’s the path to being a full-fledged financial institution of the future? Will it remain partner-reliant, or eventually take on a charter? Britt has thus far preferred the partner route, but if Chime’s deposit growth pushes partners near the $10B cap, Chime might consider buying a small bank or getting its own charter to safeguard its economics. That would be a major strategic shift, essentially transforming into a bank holding company, and it’s something to watch in the few years post-IPO. For now, the simpler approach is to add new verticals (investing, insurance, etc.) on top of the partner-bank foundation.

Chime’s future outlook is about evolution, not revolution. It has built a powerful brand and base by solving fundamental problems for everyday consumers. The next chapter involves taking those consumers further on their financial journey, helping the renter become a homeowner (perhaps one day offering mortgages?), helping the credit newbie become creditworthy (maybe offering auto loans or bigger lines), and helping the saver become an investor. All the while, Chime will aim to keep its reputation for fairness and simplicity. If it can execute, Chime could transition from being “my first bank account” for many into “my only bank account,” and perhaps even the bank for their kids, their side business, and their retirement too. That’s a long game, but Chime’s IPO will give it the capital and clout to try and make it happen. The fintech revolution it helped popularize is maturing, and Chime’s task is to prove that a mission-driven, fee-free model can mature with it, delivering not just growth, but sustainable value to customers and shareholders alike in the years ahead.

💡 TLDR: Key summary about Lovable you should know:
 
$18MN ARR in 3 Months: Lovable rocketed to $18mn ARR within 90 days of launch, with ~30k paying customers, likely the fastest ascent in European tech history. A tiny team (15 people) achieved $1M ARR per employee, shattering SaaS benchmarks.

Open-Source Launchpad: Lovable began as GPT-Engineer, a free Command Line Interface (CLI) tool that amassed 20k+ GitHub stars practically overnight. This gave Lovable an army of early adopters and 27k waitlist signups before launch, seeding its growth with zero marketing spend.

Product-Market Fit via “Reliability Unlock”: After a year of iteration, Lovable’s AI agent finally stopped getting stuck and could handle large codebases reliably. This breakthrough transformed Lovable into a must-share product, an AI that actually ships working apps. 

Best in class customer retention: Lovable boasts retention that is higher than even ChatGPT, thanks to an obsessive focus on user experience. Satisfied users became evangelists, sharing thousands of AI-built apps and driving a viral growth loop.

Community Flywheel: Lovable engineered distribution with a build-in-public playbook: A “Launched” showcase let users publish and remix AI-generated apps, gamifying sharing via contests. Partnerships with dev darlings (Supabase, Replicate) and #1 launches on Product Hunt & Hacker News further fueled awareness. All with only ~$2M spent to date.

Funding and Trajectory: After a €6.8M pre-seed in Oct 2024, Lovable raised €14.3M ($15M) pre-Series A in Feb 2025. By May 2025, over 500,000 users had used the platform. A full Series A is likely on the horizon given Lovable’s rapid revenue and user growth.

What if anyone could hire a genius software engineer on demand? Lovable is delivering on that promise. In just months, this AI-powered “full-stack engineer” went from a GitHub experiment to Europe’s fastest-growing startup, reaching an crazy $40mn in ARR within 5 months of going live.

From GPT-Engineer to Lovable

The roots of Lovable trace back to mid‑2023 when Anton Osika, then CTO of a Y‑Combinator startup and former CERN researcher, open‑sources a weekend project he calls GPT‑Engineer, a Command Line Interface (“CLI”) that can spin up an entire code‑base from a single prompt. Within weeks the repo explodes past 20,000 GitHub stars and “hundreds of thousands” of tinkerers, signalling an unmet appetite for AI that does more than autocomplete snippets.

Osika teams up with fellow Swede Fabian Hedin, who previously sold a high‑school startup, to transform the proof‑of‑concept into a product anyone, coder or not, can wield.

This open-source triumph gave Osika and Hedin instant credibility and a passionate community. By late 2023, with 27k people on the waitlist asking for more, the Swedish duo decided to take GPT-Engineer to the next level, with had audacious goal, to make software creation accessible to the 99% of people who can’t code.

They set out to build a friendlier product on top of the GPT-Engineer engine so that even non-developers could use it. The first prototype, GPTEngineer.app, quietly launched in Dec 2023 , but early versions were still rough. Over the next year, the team kept iterating through trial and error.

A turning point came by mid-2024. Lovable’s team discovered how to stop the AI agent from getting “stuck” on key tasks and enabled it to handle larger, more complex codebases. They prioritized must-have flows such as adding login, persisting data and wiring Stripe payments.

This reliability unlock, ensuring the AI could successfully plan, code, and debug full apps was crucial. With a far more capable product in hand, the founders decided it was time for a grand debut. Just two days before a November 2024 launch, they rebranded GPT-Engineer as “Lovable” , reflecting a broader vision of “software that creates software.” The relaunch was perfectly timed: Lovable’s public beta exploded out of the gate.

On November 21, 2024, Lovable’s new web app went live. It promptly hit #1 on Product Hunt (with a 5 star rating) and the front page of Hacker News. The open-source “halo” effect was evident: thousands of early fans mobilized to try it and spread the word. In its first four weeks, Lovable’s AI engineer helped users build enough projects to reach $4M in ARR. By two months in, that figure hit $10M ARR. The little side-project had officially become Europe’s fastest-growing startup ever.

Subscribe now

An AI Full-Stack Engineer That Actually Works

Lovable calls itself “the last piece of software”. An AI-powered platform that turns plain English descriptions into live applications. In practice, it gives users a chat interface to a tireless AI engineer. It is a magical experience where you simply describe your idea or features, and Lovable’s agent springs into action: scoping the project, writing code for front-end, back-end, and database, running tests, and deploying the result. Things that takes months and thousands just a couple years ago.

Under the hood, Lovable orchestrates multiple AI models and tools to achieve this autonomy. It uses advanced reasoning to create, debug, and maintain the codebase in a loop. If it encounters an error or a requirement gap, it can attempt a fix or ask the user for clarification, much like a human engineer would. This agent approach was honed from the GPT-Engineer CLI days and is Lovable’s secret sauce, enabling it to deliver production-ready software, not just snippets.

“We’ve seen hundreds of commercial apps built on the platform, at least 25,000 apps a day,” - Anton Osika

To make AI development approachable, Lovable provides a visual WYSIWAI editor (“what you see is what AI imagined”). After the AI generates your app, you can toggle into a Figma-like Visual Edit mode to tweak the UI or content directly on the page, without touching code.

This allows non-programmers to fine-tune layouts, colors, text, etc., and then save changes back to the code seamlessly. In effect, Lovable combines the speed of AI generation with the control of a traditional editor. This addresses the pain point of the difficulty of altering AI-produced code faced with other code assistants. It’s “no-code meets pro-code”: users get a smooth visual experience, but behind the scenes it’s editing code.

Finally, Lovable streamlines the deployment process. Every project comes with one-click cloud deployment, the platform integrates with popular hosting services like Netlify/Vercel and database/auth providers (Supabase, Clerk) so that a user’s app can go live instantly. In practice, when the AI finishes building your app, you can publish it to a URL with a button click (or connect your own domain). Lovable handles the DevOps drudgery: setting up repositories, deploying updates on each change, and monitoring uptime via its dashboard. This way, a non-technical creator can go from idea to a shareable, running product in an evening without tickering with local environment and manual server setup.

Hacking Growth with Community and Content

Lovable’s early growth was orchestrated through a clever build-in-public playbook that the team executed relentlessly. A cornerstone of their strategy was turning their enthusiastic user community into a self-sustaining marketing engine. One of Lovable’s inventions is “Launched”, a community app showcase at launched.lovable.dev.

Users who built something with the AI can post their app to this gallery and earn upvotes. Each week, the top-ranked apps win free usage credits, spurring friendly competition. The effects of Launched have been powerful: it gamifies sharing (hundreds of apps submitted per week), provides social proof to newcomers browsing the gallery, and even drives new users because every shared app includes an “Edit with Lovable” button, inviting others to remix it and sign up. This community content created a viral loop: users build → share → new users join → build more.

Beyond the showcase, Lovable invested heavily in fostering its community. The team runs online hackathons by partnering with VC firms like a16z and Northzone to host them, highlighting user success stories in their blog and newsletter. They also set up official Discord and Reddit channels where users swap tips and prompt ideas. All of this made early adopters feel like part of the journey. “Users became our voluntary salesforce,” Osika has said in interviews and it shows in the numbers. Thanks to word-of-mouth on X and LinkedIn, Lovable grew to 30k customers almost only through organic word of mouth.

Content marketing and strategic partnerships formed another key growth engine. Lovable’s demos are inherently shareable, watching an AI build an app in 30 seconds feels like magic, so the team leaned into short video content. They posted snappy build videos on TikTok, Twitter, and YouTube that went viral among tech circles.

Anton Osika became a hands-on evangelist, regularly sharing behind-the-scenes updates, metrics milestones, and even memes on LinkedIn and X. His informal, authentic style stood out in a space normally dominated by polished product shots. This authenticity helped him amass a following and kept Lovable constantly in the social media discourse.

Lovable also piggybacked on larger platforms and communities. They forged integration partnerships with popular developer tools. For example, doing joint webinars with Supabase (database) and Replicate (AI model hosting) to show how Lovable plugs into those services. Each collaboration cross-pollinated user bases (“Supabase’s community discovered Lovable, and vice versa”), adding credibility by association. The Lovable team even joined an official Andreessen Horowitz AI showcase alongside bigger-name startups, boosting their visibility among tech influencers. And of course, the Product Hunt launch was executed flawlessly: the team rallied their open-source fans to upvote en masse, yielding a #1 Product of the Day and a flood of sign-ups. In short, Lovable created a flywheel where community, content, and product excellence reinforced each other. A textbook example of scrappy, modern go-to-market execution.

Hyper-Growth

The numbers behind Lovable’s rise are eye-popping. Three months from its public launch, Lovable reached $18M in ARR, with 30,000 paying customers on board. For context, it took Slack about 3 years to hit similar revenue. Lovable did it in a single quarter, without a sales team.

This translates to roughly $1M in ARR per employee, given the team had just 15–18 people at the time, literally off the charts for a SaaS startup. For a software company with $20–50M ARR, a ‘great’ ARR/employee is ~$275k. Lovable’s ~$1M is wild. This efficiency is a testament to the product-led growth and the leverage of having AI handle much of the heavy lifting.

User adoption has been equally impressive. By Feb 2025, over 500,000 users (free and paid) had signed up to try Lovable’s AI builder. Many churned after tinkering, but the core paying base kept growing and using the product. Lovable reported that by month three, users were collectively creating 25,000+ new apps per day on the platform. This hints at strong engagement and that people are incorporating Lovable into real projects and workflows.

Another metric pointing to product-market fit: Lovable achieved 85% 30-day retention of users during its early growth, surpassing even OpenAI’s ChatGPT retention. In other words, the majority of users who built something with Lovable once were coming back to build more, a sign of genuine value, not just hype.

On the funding front, Lovable has been strategic. The founders raised a €6.8M pre-seed (led by Hummingbird and byFounders) in late 2023 to assemble a small team of elite engineers and cover initial compute costs. Notably, they held off on a bigger raise until they had clear traction. Only after hitting ~$10M ARR did Lovable announce a $15M pre-Series A in February 2025.

Creandum led the round, with prominent angels like Charlie Songhurst (ex-Microsoft), Adam D’Angelo (Quora), and Thomas Wolf (Hugging Face) joining. The capital is earmarked to expand integrations (e.g. deeper ties into tools like GitHub, Supabase) and scale infrastructure for the surging user base. Impressively, Lovable had only spent about $2M of its pre-seed money by the time of this raise , showing high capital efficiency. With fresh funds, the team is likely growing beyond 18 people, but still pre-Series A, they maintain a lean, builder-heavy roster, about half the team are engineers, and most others are ex-founders or high-agency operators. This headcount discipline has helped Lovable set a new bar for ARR per employee, and gives them a long runway to keep executing at lightning speed.

Lovable and the AI Coding Wars

Lovable operates at the intersection of no-code tools and AI coding assistants, carving out a unique position. It’s useful to map the landscape of “AI coders” to see where Lovable stands. On one axis, consider the target user: professional developers vs. non-programmers. On another axis, consider the AI’s role: assisting a human vs. autonomously building. In Lovable’s case, the target user is the non-programmer (or any builder without deep coding skills), and the AI’s role is agentic and autonomous, making it a distinct offering in the market.

Direct competitors (AI app builders): A few startups are pursuing a similar vision of one-click app creation. Bolt.new is one notable competitor, an AI-powered app builder that also generates full-stack code from prompts. Like Lovable, Bolt can scaffold entire web apps and deploy them, though it has stayed more focused on developers as the user (it pitches itself as a way for devs to speed up building standard apps). So far, no competitor in this category has matched Lovable’s momentum or reliability, but the space is young. However, we’re likely to see more challengers given the massive TAM (virtually every knowledge worker with an idea could be a customer).

Adjacent competitors (AI coding assistants): On the developer-focused side, tools like Replit, Cursor, and Windsurf (formerly Codeium) are tackling parts of this problem. These products integrate AI into coding environments to help engineers code faster. For example, Replit’s Ghostwriter can autocomplete code and even generate entire projects from a prompt, but it operates within a developer IDE and assumes the user can refine the code. Cursor is an AI-enhanced code editor that uses an agent to manage your files and implement requests (similar underlying tech to Lovable, but aimed at programmers working in Python/TypeScript projects). Windsurf/Codeium has introduced an “AI agent” in its IDE that can run build commands and fix errors automatically, again oriented to coders. These tools live in the world of software professionals, they can supercharge productivity, but they are not turnkey app creators for a non-coder. As TechCrunch aptly noted, platforms like Cursor and Copilot “are mostly confined to a world of trained engineers,” whereas Lovable opened up app-making to anyone.

Big tech and incumbents: The specter of competition from the giants looms as well. OpenAI’s own Codex model (and GitHub Copilot powered by it) provides the core AI brains for many of these tools. So far OpenAI hasn’t rolled out an end-user product that builds apps from scratch, instead they supply the API (e.g. GPT-4.1) that Lovable runs on. Microsoft (which backs OpenAI and GitHub) has integrated Copilot deeply into Visual Studio and even into Power Platform for low-code scenarios. We may soon see an offering that tries to do what Lovable does (for instance, one could imagine ChatGPT integrating a “build me an app” plugin or Microsoft releasing a Copilot Studio for non-programmers). For now, Lovable’s head start and focus give it an edge: it’s a standalone, platform-agnostic solution laser-focused on AI app generation, while the big players are either enablers (OpenAI) or tied to their ecosystems (Microsoft’s copilot in Azure/Office).

Lovable’s edge: Speed and focus have been critical. By the time incumbents or well-funded startups noticed this opportunity, Lovable had already accumulated a 500k-strong user base and, crucially, a dataset/feedback loop from tens of thousands of real app builds. The reliability of its agent is a differentiator frequently cited by users, “Lovable is the best way to get something that actually works,” Osika argues, contrasting it with code assistants that often leave users with half-baked output. Moreover, Lovable’s community-driven moat (templates, showcase, active discourse) means it’s not just an AI model you can copy, it’s a network of builders and shared solutions. In effect, Lovable is attempting to build a network effect: the more people build and share apps on the platform, the more value new users get (via examples, templates, community help). This could prove a lasting advantage against tech giants, who have scale but not that grassroots developer-love. However, Lovable will need to keep innovating aggressively to maintain its edge, as the AI coding landscape is evolving daily.

Risks and Challenges

Despite its sensational start, Lovable faces a set of significant risks moving forward:

• Sky-High Compute Costs: Generating full applications with AI agents is resource-intensive. By Osika’s own admission, Lovable likely has “the biggest spend on LLMs among startups in Europe” at this point (reflecting huge usage). Running GPT-4, Claude, and other models for thousands of users could burn cash quickly, especially if users on fixed monthly plans heavily utilize the service. The company’s gross margins may suffer if AI API costs don’t come down. In the long term, Lovable might need to develop its own optimized models or inference infrastructure to rein in costs, or lean in more on usage-based pricing to balance heavy users.

• Code Quality and Liability: While Lovable strives for reliable outputs, AI-generated code can still introduce bugs or security vulnerabilities. If a Lovable-built app fails catastrophically or gets breached, who is responsible? The user or the platform? There’s also intellectual property risk: the AI might inadvertently produce code snippets that resemble licensed code, raising legal questions. Lovable will need to continuously validate and sandbox generated code. They’ve partially addressed this by open-sourcing the underlying engine (creating transparency) and allowing users to own the code (the user’s repo), but as usage scales, quality control will be an ongoing challenge. Enterprise customers in particular will demand strong assurances on security and compliance.

• Incumbent Bundling and Competition: The flip side of being a pioneer in AI app-generation is drawing attention from industry giants. It’s easy to imagine players like Microsoft, Amazon, or Google bundling similar AI builders into their cloud platforms or no-code tools. For instance, Amazon could integrate a “GPT app builder” into AWS for its customers. If such offerings come free with a larger service, Lovable might face pressure to differentiate. Additionally, traditional low-code/no-code platforms (e.g. Webflow, Bubble) are starting to add generative AI features – they already have large user bases and could quickly catch up on the AI front. Lovable must move fast to establish an ecosystem (and maybe forge partnerships with the big clouds rather than compete head-on).

• Sustainability of Hype: Hyper-growth can be a double-edged sword. Lovable’s astonishing early metrics set sky-high expectations. If ARR plateaus after the initial rush or early adopters churn out once the novelty wears off, the narrative could quickly shift. The company will need to move beyond the tech enthusiast niche into mainstream or enterprise use-cases to sustain growth. That involves refining the product (for even non-techier users), providing support/education, and perhaps curating successful use cases in specific verticals. Essentially, Lovable must prove it’s not just riding a hype wave but is a durable business. Maintaining its developer community goodwill while also courting enterprise clients will be a balancing act.

Next Act: What’s Ahead for Lovable

As of May 2025, Lovable stands at an inflection point. Flush with new funding and a record-setting launch behind it, the startup is charting its next moves to solidify its lead and expand its platform:

• Enterprise & Teams Focus: An obvious next step is building out an enterprise-grade offering. Large companies may want to use Lovable’s AI internally, but will need features like on-prem or VPC deployment, single sign-on (SSO), audit logs, and the ability to restrict the AI to proprietary data. Lovable has already started an Enterprise tier with options to “opt out of data training” and dedicated support. We can expect a full enterprise SDK or API that lets development teams integrate Lovable’s engine into their own IDEs and workflows. This could open a new revenue stream (selling to IT departments or SaaS companies) and make Lovable more defensible by deeply embedding it in organizations.

• Marketplace & Ecosystem: With thousands of apps and components being generated, Lovable may create a marketplace for templates, plugins, and AI extensions. Imagine new users being able to start from a community-contributed template (an e-commerce app, a CRM system, etc) instead of from scratch. In fact, the Launched showcase and prompt library are early signs of this. A formal marketplace could let creators sell polished templates or custom add-on modules (for example, a plug-in that adds a Stripe payments integration or a specific machine learning microservice) to others. This would further drive the flywheel: top community builders can monetize their expertise, and new users get faster time-to-value. It also positions Lovable as not just a tool but a platform others build on, increasing lock-in and network effects.

• Continued Product Innovation: On the product roadmap, the team is likely tackling even bigger challenges: multi-modal inputs (e.g. generating apps from a hand-drawn sketch or from a competitor’s website URL), more complex application types (mobile apps, not just web; perhaps desktop software), and deeper customization for advanced users. The core AI model loop will also improve with data, every failed generation or user edit is feedback to make the next output better. There’s talk in AI circles about fine-tuning models for “software agent” tasks; Lovable could train its own model or a GPT-4 fine-tune on the trove of successful app builds it has, further improving reliability and speed. We might also see AI pair-programming features within Lovable, eg the AI suggesting improvements continuously as a user manually edits code – to cater to semi-technical users.

• Long-Term Defensibility: To succeed in the long run, Lovable will need to cement advantages that competitors can’t easily copy. One such advantage is data, Lovable is amassing a unique dataset of prompt-to-app transformations and user feedback on those apps. If they leverage this to continuously fine-tune their AI, they could develop a proprietary model that outperforms general models for coding tasks. Another moat is the community and brand: Lovable has become synonymous with AI app-building, especially after the virality and media coverage. That mindshare, combined with the active community and content machine, is hard to replicate. Finally, Lovable could cultivate an ecosystem lock-in: if businesses start relying on Lovable for critical internal tools (and perhaps build custom integrations around it), switching to a competitor or DIY solution becomes non-trivial. In summary, the next act for Lovable is about evolving from a lightning-in-a-bottle startup into a sustainable platform – extending its lead in technology, broadening its user base (while keeping early adopters happy), and navigating the inevitable challenges that come with rewriting the software development playbook.

Imagine this scene: a traveler steps off a 13-hour flight to Bangkok, exhausted and disoriented. But before they can mumble "where can I buy a SIM card?", their banking app chimes with a series of notifications:

"Welcome to Thailand! You're now connected with 2GB of data. Enjoy your trip!"

"Current exchange rate: 1 USD = 32.5 THB. Would you like to convert $200 to ฿6,500 at this rate?"

"Your card is ready to use. We'll automatically show prices in USD while you shop."

No hunting for airport kiosks. No swapping tiny SIM cards with jetlagged fingers. No seeking out currency exchange counters with questionable rates. Their banking app quietly handled everything mid-flight, creating a seamless arrival experience.

This isn't science fiction, it's happening right now across the fintech world. An unlikely convergence is underway as digital banks aren't just handling money anymore, they're keeping phones connected too. From Revolut in Europe to Nubank in Brazil, fintechs are aggressively jumping into the telecom business through eSIM offerings, creating an integrated travel experience that combines connectivity and currency in one place.

It raises a delightful question as travelers breeze through immigration: why on earth is my fintech app also my phone company?

The eSIM Revolution: Why Now? 📱

This whole convergence rides on a technology that's been quietly changing our phones: embedded SIM (eSIM).

Unlike the physical SIM cards we've fumbled with for decades, eSIMs are built directly into devices. Most recent iPhones and Android phones already have them, Apple even eliminated physical SIM trays entirely in US iPhone 14 models and beyond.

The numbers reveal a technology hitting its tipping point:

• Nearly 600 million eSIM-equipped smartphones exist worldwide

• Travel eSIM market size is projected to explode from around 2 billion today to 260 billion by 2034

But technology alone doesn't explain the timing. What's changed now is the emergence of "telecom-as-API" platforms. Companies like 1GLOBAL and Gigs have built infrastructure that handles all the messy backend work of telecom operations. Ten years ago, a bank offering mobile service would have been technically impossible due to enormous infrastructure barriers. Today? It's basically copying and pasting an API key.

This is the AWS moment for telecoms. Just as cloud computing made every company potentially a software company, these APIs are making every app potentially a telecom provider.

The friction is gone. The playground is open.

Why eSIM Is Fintech's Perfect Match 🔍

Fintechs have tried expanding into many adjacent spaces, from insurance to BNPL loans to carbon footprint tracking. So why has connectivity emerged as the clear winner? Three strategic factors explain this convergence:

1. Beefing up the Subscriptions plan 📦

Fintechs have long sought ways to deepen engagement and lock in recurring revenue. Over time, the subscription model has emerged not just as a monetization tool, but as strategic infrastructure, anchoring everything from premium cards to cross-border payments to insurance.

Mobile data plans fit neatly into this ecosystem. They're inherently recurring, immediately useful, and highly valued by globally mobile users. Embedding connectivity into financial subscriptions adds a tangible benefit that users instinctively understand, reinforcing the value of premium tiers.

The alignment between banking subscriptions and connectivity needs makes eSIM offerings a natural extension. A user who pays monthly for financial services is already primed to expect ongoing value; throwing in "3GB of global data monthly" makes the offer more concrete and compelling than nebulous perks like "enhanced fraud protection." In this way, mobile plans don't just complement the subscription model, they amplify its effectiveness.

2. The Travel Flywheel Effect ✈️

Fintechs have been assembling a "travel stack" for years:

• Multi-currency accounts for forex

• Premium travel insurance

• Airport lounge access

• Hotel and flight booking integrations

Adding eSIM is a puzzle piece that makes the whole machine spin faster.

Picture the journey: A traveler books flights through their banking app's deals, accesses the airport lounge with their premium card, lands with instant connectivity from the same app, then uses that app's card for local purchases. Each service feeds the next in a virtuous cycle, a flywheel generating more engagement, more data, and more revenue.

3. Smarter Security Through Location 📍

When your banking app provides your connectivity, it gains powerful contextual data: where you are, when you’re active, and your usage patterns.

This location intelligence is a game-changer for fraud prevention. For example, Monzo’s “Known Locations” feature allows users to register trusted places, like home or office to authorize sensitive transactions based on GPS location. But with an integrated eSIM, fintechs can level this up even further. eSIMs provide reliable, carrier-verified location data, resistant to GPS spoofing or device tampering.

If a card is used in Paris but the user’s phone, confirmed by network-level connectivity, is clearly in São Paulo—that’s an instant, accurate red flag. Conversely, verifying that the device and card are present in the same country via cellular networks allows fintechs to significantly reduce false positives, providing a smoother, frictionless experience for legitimate users.

This direct access to trusted location signals eliminates reliance on third-party data, tightening security protocols and reducing unnecessary friction. It’s a win-win: stronger fraud protection for fintechs, and seamless experiences for users traveling abroad.

Subscribe now

Global Pioneers: Who's Leading the Charge 🌎

Revolut: From Banking App to Full-Service Carrier

The UK-based fintech was among the first to integrate global eSIM. For about $55/month, Ultra plan subscribers get premium banking features plus 3GB of global data monthly, easily $30-50 worth of roaming data included "for free." According to its latest annual results, 600k customers purchased an eSim plan with the fintech in 2024. Not trival numbers at all.

But Revolut isn't stopping at data-only eSIMs. They're now expanding into full mobile plans with calls and texts in the UK and Germany, essentially becoming a full-fledged carrier. Their strategy is clear: test new markets with eSIM data plans first, then launch complete mobile services once traction is established.

The strategy is working. Revolut saw subscription revenue climb 74%, with paid subscribers surging 45%. Perhaps most tellingly, eSIM became their #1 non-banking product by usage. That's not just a feature, it's a growth engine.

Nubank: Connectivity as Premium Incentive

Latin America's largest digital bank has taken a different approach, focusing on using connectivity to drive premium account adoption.

Their Ultravioleta cardholders get a free 10GB roaming data eSIM annually, covering 195 countries. In a region where roaming data costs are prohibitively expensive, this perk carries significant value.

Nubank has created interesting synergies with their "NuCel" product. Beyond connectivity, customers gain access to an exclusive savings account with a 120% CDI return (substantially above typical rates) on deposits up to R$10,000 (US$1,748), guaranteed for at least one year.

This approach demonstrates how connectivity can be leveraged not just as a standalone feature, but as an entry point to a broader ecosystem of premium financial services.

The Roadblocks: Not All Smooth Roaming ⚠️

Fintechs diving into telecom is an exciting frontier, but there's more to this story than flashy announcements suggest. Revolut and Nubank's mobile connectivity plays face real challenges that make this marriage more complex than it first appears:

1. MVNO Dependency: Fintechs rely on established telecom providers (e.g., Vodafone, Telefónica, Verizon) for network access, which leaves them vulnerable to pricing pressure and limited control over technology and quality of service. Strategic negotiations and multi-carrier partnerships can mitigate, but not fully eliminate, this dependency.

2. Infrastructure Economics: And if fintechs are thinking of going direct, it then requires enormous upfront investments, billions in spectrum licensing alone, plus substantial network infrastructure buildouts. While this high barrier protects fintechs from extensive competition, it also permanently ties them to partnerships rather than independent control.

3. Operational Complexity: The minute you offer connectivity, you inherit telecom-level support expectations. "My internet won't connect in rural Vietnam" is a very different problem than resetting a card PIN. Poor service could damage the fintech's core brand.

The Next Frontiers: Where This Convergence Leads 🔮

As fintech-telecom fusion continues evolving, several exciting possibilities emerge:

Full Voice/SMS Integration: More fintechs could follow Revolut's lead into full-service mobile plans, complete with phone numbers and calling. This edges closer to complete convergence, more bank app might assign you a phone number in the near future.

Context-Aware Financial Services: The marriage of location and financial data enables predictive features. Imagine landing in Tokyo and a banking app immediately asks: "Want to convert ¥50,000 at today's rate?" Or it notices a booked flight to London next week and offers: "Shall we pre-load £200 and activate your UK eSIM?" This moves banking from reactive to proactive.

The End of Industry Boundaries 🧩

This isn’t just fintech trying on a new costume. It’s a genuine shift in what users expect from the apps that hold their money, and now their mobile identity. Banking and connectivity are fusing, not for novelty’s sake, but because it solves real friction at the exact moment users need it.

The best fintechs aren’t chasing telecom margins. They’re embedding connectivity to drive retention, improve risk, and create magical moments, like landing in a new country with your phone and money working in sync.

What used to be siloed services are dissolving into single, elegant experiences. The question isn’t “why is my bank a telco?” It’s “why wasn’t it before?”

In each edition you’ll get:

• Game Plan: What is the Strategy?

• The Recipe: How did it work?

• Deep Dives: Where can I find out more?

Think of it as your go-to source for compact strategy insights, landing directly in your inbox.

Subscribe now

Alright, let us get into it!

Game Plan

Strategy in a Nutshell

In 2002, a peculiar shoe called the Beach made its debut. Two decades later, that same design would become the foundation of a $5 billion empire. This is the story of Crocs and their strategy which challenges the conventional wisdom about product strategy and brand building.

The journey wasn't always smooth. During the 2008 financial crisis, Crocs was on the edge of bankruptcy. The following years saw stagnant growth as the company grappled with an identity crisis. By the early 2010s, Crocs had expanded to 230 different styles – from sandals to high heels – diluting the very thing that made them unique: their iconic clogs.

This cautionary tale of expansion nearly spelled disaster. But sometimes, the path to growth isn't about doing more – it's about doing less.

💡 Strategy Playbook: Streamline and perfect your core product.

The Recipe

Three Ingredients for Success

1. Focusing on the Crown Jewel

In the late 2010s, Crocs launched a ruthless strategy to focus on its core product, cutting half of the existing SKUs. Gone were the boots, the sandals, and those experimental high heels. The strategy? Zero in on what made them famous (or infamous) in the first place - the iconic clogs.

They correctly identified the potential of their iconic clog and focused Crocs strategy around that product.

2. Sell an idea, not the product

With laser focus on their core product, Crocs took their iconic clog and transformed its narrative completely. Instead of fighting the "ugly" label, they embraced it. The clog became more than just footwear but a statement about confidence and self-expression.

The existing Jibbitz played perfectly into this reinvention. These little charms that had been around since the early days suddenly gained new meaning. They became more than just decorations but tools for personal storytelling. In a world where everything looks the same, your Crocs could be uniquely yours, decorated with symbols of your personality, interests, and style.

3. Amplifying through collabs

The final piece of the revitalization is to “re-wire” peoples’ brains and change the image of Crocs brand is people's mind.

To do that, Crocs harnessed the power of collaborations. This allowed Crocs to benefit from the fame or cultural relevance present within the collaborator, such as a celebrity or another iconic brand.

Justin Bieber, Post Malone, Balenciaga and even KFC got in on the action. Each collaboration felt authentic because the partners actually got their hands dirty in the design process, creating their own Jibbitz and putting their genuine spin on the shoes.

Through these strategic partnerships, Crocs engineered a cultural shift that transformed their narrative from being "ugly and uncool" to "ugly but very cool and trendy". What started as a practical comfort shoe became a canvas for cultural moments, each collaboration adding a new layer to Crocs' evolving story.

Bottom Line:

The magic? These three moves complemented with each other. The focus on clogs gave them clarity. The reinvented narrative of self-expression gave people permission to wear them proudly. And the collaborations? They transformed Crocs from a comfortable shoe you'd hide in your closet to one you'd proudly wear on Instagram. That's how you turn a near-bankruptcy story into a billion-dollar comeback.

Deep Dives

Resources for the Curious Minds

1. Yann Le Bozec, Senior Marketing Director, Crocs - How Crocs changed consumer perceptions

The revitalization strategy that changed Crocs, straight from the team leading it.

2. Crocs’ key pillars of growth - Q3’24 investor’s presentation

And that’s it for this week!

If you enjoy my post, please consider sharing it or subscribing, see you in the next piece!

Share

Subscribe now

— Anthony 🚀

In each edition you’ll get:

• Game Plan: What is the Strategy?

• The Recipe: How did it work?

• Deep Dives: Where can I find out more?

Think of it as your go-to source for compact strategy insights, landing directly in your inbox.

Subscribe now

Alright, let us get into it!

Game Plan

Strategy in a Nutshell

"Fun" isn't just for games—it's serious business. Companies are pouring money into figuring out how to make their products more enjoyable. Why?

Because fun creates a powerful cycle: happy users talk, and their chatter attracts more users, which then spurs more subscribers to your application. These additional proceeds can then be used for more research into making the app even more fun.

It's a self-feeding loop that's as profitable as it is delightful.

Duolingo took this "fun formula" and sprinkled it into something traditionally not-so-fun: learning.

They didn't stop with the app; they infused fun into their entire brand. But how did they turn a task as daunting as language learning into something so fun?

Let's dive in.

💡 Strategy Playbook: Incorporating fun into your strategy pays huge dividends.

The Recipe

Three Ingredients for Success

1. Learning Gamified

Duolingo recreates fun in its app by borrowing a page from the most addictive playbooks out there: video games.

They transformed the app into a digital playground where learning feels more like playing.

Through gamification, they pull out some of the successful elements and dopamine hits from the world of gaming. Here are some examples:

• Experience Points (XP): We like to gain things 

• Streaks: We do not like to lose things

• Leaderboards: Competition 

  

This keeps learners engaged and motivates them to spend more time learning.

Fun stat here: Duolingo has >3 million active users with streaks of more than 365 days 🤯

2. Unique Branding Campaigns

But the fun doesn't stop with the app. Duolingo's branding campaigns are a masterclass in creativity, reaching out and grabbing attention in the most unexpected places.

They focus on organic marketing and tap into trends and cultural zeitgeist which allows Duolingo to jump onto trending topics (Trendjacking) and become part of the conversation.

Some examples of their partnerships to promote their brand:

1. HBO, Game of Thrones: Launched a High Valyrain language course when the House of the Dragons series was released

2. Peacock, Love Language: Reality TV dating show where none of the contestants speak the same language

3. CBS, Star Trek: A Klingons language course for Star Trek fans

4. Netflix, Emily in Paris: Offering anyone named Emily a free month of premium membership

When people sign up for fictional language courses, 43% of them begin to study another language, allowing Duolingo to build not only a fun brand but also acquire more users.

3. Unhinged Marketing

When it comes to their social media content, Duolingo went beyond fun and became completely unhinged.

Their social media strategy focuses on:

1. Focusing on entertaining instead of selling

2. Hopping on trending topics 

3. Engaging storytelling:

   • Ongoing feud with Google Translate

   • Obsession with Dua Lipa 

   • etc…

This comes across as authentic and relatable to its audience, especially the younger demographics, and also massively captivated their followers on social media. (And also an absolute dream job for their marketing department folks…)

Bottom Line: The Economics of Fun

Making the whole Duolingo ecosystem fun creates incredible benefits for the economics of the business. With increased brand engagement with its multiple channels of fun and novelty, the cost of acquiring new customers is reduced dramatically.

About 80% of their users are acquired organically through word-of-mouth virality, which is incredible.

Deep Dives

Three resources for the Curious Minds

1. TED Talk by Duolingo Founder, Luis Von Ahn, on making learning addictive

“I hope for a future where screen time is not a bad thing, in which we can deliver high quality education to everyone, rich or poor, using a mobile phone.”

2. Duolingo S-1 (pre IPO) form that dives deep into their business mode

A deep dive into the Duolingo’s business model.

3. Discussion on the gamification on the Duolingo platform

Particularly interesting is how Duolingo has features that satisfy common player types: Achievers, Explorers, Socialisers and Killers.

And that’s it for this week!

If you enjoy my post, please consider sharing it or subscribing, see you in the next piece!

Share

Subscribe now

— Anthony 🚀

In each edition you’ll get:

• Game Plan: What is the Strategy?

• The Recipe: How did it work?

• Deep Dives: Where can I find out more?

Think of it as your go-to source for compact strategy insights, landing directly in your inbox.

Subscribe now

Alright, let us get into it!

Game Plan

Strategy in a Nutshell

Grab was once dubbed the Southeast Asian Uber, kicking off its journey in 2012 thanks to Anthony Tan (not me) and Tan Hooi Ling.

So, when the real Uber decided to step into the Southeast Asian ring in 2013, many expected Grab to be quickly knocked out. But,nope! After a saga that lasted more than half a decade, Grab stood victorious, with Uber selling their SEA business to them. 

💡 Strategy Playbook: Tailor your product to the market… if not someone else will.

Here's how they did it.

The Recipe

Three Ingredients for Success

1. Get hyper-localised

Southeast Asia has a vibrant tapestry of cultures, with over 1,000 native languages and a non-homogenous market.

Grab recognises this and tailors regional rollouts instead of copy and pasting strategies. True, it is not a scalable strategy, but it is a winning one. Here's a peek at how Grab tailored its strategy:

💵 Cash is King: Recognizing the cash-dominant economy of Southeast Asia, Grab welcomed cash payments right from the start in 2012, while Uber played catch-up three years later.

🛵 Two Wheels over Four: In the traffic-jammed streets of Southeast Asia, motorcycles zip through faster than cars. Grab introduced GrabBike in 2014, giving them a two-year head start over Uber.

By focusing on what users really needed rather than banking solely on their brand, Grab raced ahead.

2. Strategic Alliances

In the world of startups, cash might be king, but the right partnerships are important too.

Capital is a commodity and Grab didn't just look for investors; they sought allies who brought more to the table than just money.

By aligning with local power players across the regions they entered, Grab secured not just funds but invaluable local insights and licenses, a stark contrast to Uber's strategy of relying on American dollars and expertise.

From government-linked Vertex Ventures in Singapore to giants like DiDi in China and Toyota in Japan, Grab's investor roster reads like a who's who of regional influencers.

3. Go Hard & Go Fast

Grab moved fast in their early years. They expanded their core ride-hailing service across six countries within three years of launch. But they didn't stop there. 

They quickly diversified into verticals like payments and courier services, always staying one step ahead of Uber.

This strategy of rapid expansion and diversification positioned Grab perfectly to capitalize on any missteps by their competitors.

When Uber stumbled in 2017 amidst a whirlwind of scandals, Grab was in the perfect position to deliver the final blow by acquiring Uber's Southeast Asian business. 

Deep Dives

Three resources for the Curious Minds

1. In-depth documentary of Uber vs Grab

"They were 24 by 7 focused on the regional market, they had people on the ground, they had engineering team on the ground and their decision making could be a lot faster than ours just because of the distance"

2. Grab Pre-SPAC investor presentation here

A deep dive into the Grab business model, with hyperlocalisation taking up a full section

3. The landmark deal that signals the defeat of Uber in SEA - here

Not a complete failure for Uber as they take a 27.5% stake in Grab, allowing them to reap the benefits of Grab's growth

And that’s it for this week!

If you enjoy my post, please consider sharing it or subscribing, see you in the next piece!

Share

Subscribe now

— Anthony 🚀

If you haven’t subscribed yet, join the other folks interested in startups by subscribing here:

Subscribe now

Hello everyone!

VISA is one of those companies where people think they know what they do but cannot really explain what exactly they do. Myself included.

This piece will be part of a 3 part series of my summary of the fantastic episode by the Acquired podcast on Visa. As someone in the payments industry, writing this summary has been great fun for me, and I'm excited to share these insights with you.

This is part I of the Visa story:

• Part I: The Seedling of Visa

• Part II: The Birth of Visa

• Part III: The Future of Visa and Payments

Let’s dive right in!

Visa Background

Visa is an important connector in the world of global financial transactions, extending far beyond its common perception as merely a credit card issuer. Visa is like an expansive, secure highway, facilitating swift and safe monetary transfers worldwide, whether it’s for a simple coffee purchase or a lavish international vacation. 

At the heart of its operations, it provides a robust network that financial institutions utilize to offer Visa-branded cards. Each transaction made with a Visa card involves a modest fee paid by the bank to Visa, going back to the previous highway example, it collects a toll on every vehicle going through their robust highway. 

This toll road business model is an incredible one for Visa, it has the 11th largest market cap in the world and operates at a net profit margin of ~53%, far ahead of any company of similar size. 

The US Payment Industry History

To fully grasp Visa's ascent, we need to rewind to an era where cash and cheques were the norm. Each cheque presented its own set of challenges – costs for processing and a time lag that was far from ideal.

Eventually, merchants figured that instead of moving money daily, how about I just leave a tab for my customers and settle at the end of the month? And so They introduced charge accounts, enabling customers to accumulate charges and pay them off monthly. This not only streamlined transaction clearing to once a month but also significantly cut down operational costs.

Initially, these accounts were exclusive to individual stores. But soon, corporations like Standard Oil started to integrate the accounts across all their branches sending 250,000 unsolicited cards to all their customers for their charge accounts.

However, this system still isn't ideal for consumers and merchants alike. Imagine consumers lugging around dozens of cards, one for each store – hardly practical. Merchants, on the other hand, were navigating in the dark regarding the credit history of their customers with other establishments.

This unstable state pushes for a network to be formed, one where retailers get together and have a standardized charge account system.

Yet, this network was not without its ironies. By participating in this unified network, merchants inadvertently supported their competitors. Imagine a local bookstore and a national chain sharing the same payment platform. Each swipe of a customer's card at the local store indirectly reinforced the network that included the rival chain, effectively equipping their competition with the same transactional advantages. This paradox highlighted a fundamental tension in the new system, that collaboration for convenience at the cost of potentially aiding one's market rivals.

It is here where players such as Diners Club and American Express first arrived on the charge card scene.

Early players: Diners Club/American Express

Both Diners Club and American Express started out by targeting a specific group. 

Diners Club, for instance, zeroed in on the restaurant industry, establishing a shared charge account system among restaurants. This strategy is shrewd, in the competitive but non-exclusive restaurant landscape, no single establishment expected lifetime loyalty from a diner. This approach facilitated an easier buy-in from restaurants, allowing them to collectively reap the benefits of the network without fearing the loss of their unique customer base.

Diners Club's model proved successful enough to levy charges as high as 7% on merchants, and they even implemented card fees for users. Yet, despite these fees, sustaining profitability was challenging. The cost of maintaining such a network, especially with the limited technology and the heavy reliance on manual processes, was substantial, and this is just within the confines of New York City.

American Express, observing Diners Club's success, decided to enter the fray. They launched a charge card program tailored for the business traveler, expanding the concept beyond the realm of dining and into broader commercial travel and also marking the entry of a bank-like entity into the scene.

The BankAmericard

Make me some loans

As charge cards gained traction, Bank of America spotted the perfect moment to revolutionize the market with their twist: introducing credit. This strategic move aimed to elevate their services in several ways:

1. Streamline their lending programmes

2. Make more money by giving out loans

Unlike charge cards, which required full monthly payment, this credit card allowed customers the flexibility to convert their spending into loans, offering a more streamlined, on-ramped approach to introducing consumer loans.

The credit card program didn't just increase the number of loans issued, it fundamentally changed the dynamics of money flow within the bank. Having credit programmes minimizes the amount of money actually leaving Bank of America. 

Traditionally, loans involved a tangible cash exchange, the bank handed out cash, which the consumer spent, and the merchant eventually redeposited. This process, while straightforward, meant the actual cash left the bank's hands.

With the BankAmericard, this cycle was streamlined. The consumer goes directly to the store, pays with the credit card, transactions get authorised, and if merchants banks with Bank of America, no money leaves Bank of America, it is just a simple ledger update. This frees up capital and float for the bank.

And so they mailed almost 65,000 cards indiscriminately with the same credit limit to every customer.

Give me back my money

And of course, fraud happens. 

The credit card loans are practically unsecured lending, and when you issue thousands of unsecured credit lines, as they did indiscriminately, you're practically inviting fraud.

There was $20 million of fraud within the first pilot program and 22% of the credit that they issued ended up being default or delinquent, which is five or six times of what their delinquency rate was before on traditional lending. These were daunting figures, enough to make any financial institution reconsider its strategy.

But Bank of America was undeterred. They saw the untapped potential in the credit card market and were prepared to weather the storm.

Their conviction paid off. Within three years, they not only brought fraud under control but also turned the program profitable. This success, however, was closely guarded – a secret strategy that kept competition at bay. As a result, within the first six years of the BankAmericard, only a handful of new credit cards emerged.

By being first to the new market of credit card, Bank of America can operate within a zone of uncertainty for many years. Competitors are looking to them to see whether this credit card thing works before launching their own. By keeping the profitability a secret, Bank of America is able to extend this zone of uncertainty build their moat, and sign up more merchants and consumers, securing their position as pioneers in a then-uncertain market (AKA productive uncertainty).

BankAmericard Network

Eventually, Bank of America let the cat of out the bag with the launch of the BankAmericard service organisation to license out their credit card program and network to banks across the country.

Licensee/franchisee would pay Bank of America a franchise fee and also a percentage of their transaction value. Within two years, a couple of hundred banks had signed up with six million cardholders across the country and beyond the country. 

However, there was a problem. 

Before franchising its network, BankAmericard usually operated within a closed-loop network. All parties of the transaction were their customer. There are no differences in the: 

• Banks of their consumer's card (Issuing Bank)

• Bank of the merchant (Merchant/Acquiring Bank)

With the franchising, however, this was rarely the case.

The issuing bank and merchant banks are not the same party. This then transforms into an open-loop network. With the open loop network, there is a need for an interchange between different parties to enable systems to talk to each other and settle these transactions. 

But here's the catch. Bank of America's network was designed for a closed loop, and they weren't equipped to address these complexities for their licensees. In their view, franchising was more about marketing than infrastructure development.

So the franchisee has to go back in time to how cheques used to work.

Merchants accept BankAmericards and collect invoices. At the end of a month, the acquiring banks would then take the invoices and individually mail the issuer banks to get the money. But this is all great for Bank of America, it’s just internal number movement and they have no incentive to change this.

Yet, it was precisely this inefficiency that planted the seeds for the emergence of Visa.

And that’s it for this week! In the next part, we will dive into The Birth of Visa. Stay tuned!

If you enjoy my post, please consider sharing it or subscribing, see you in the next piece!

Share

Subscribe

— Anthony 🚀

If you haven’t subscribed yet, join the other folks interested in starting your own company by subscribing here:

Subscribe now

Actionable Insights

I think that Midjourney is still early in its founding and it might be premature to call them a success. But, I was curious how a bootstrapped AI company managed to lead the pack, hence this article!

Here are key takeaways I have learned from the decisions that Midjourney has made at the beginning of its journey.

💡 Focus on Core Resources Over Capital: Recognize what your key resource is. Holz recognized early on that the key resource for Midjourney wasn't money but computing power. This strategic focus demonstrates the importance of identifying and leveraging the most crucial resources for your business, beyond just financial capital.

🌐 Leveraging Community and User Feedback: Midjourney's integration with Discord showcases the power of community engagement and rapid user feedback in refining a product. This approach led to a vibrant, self-growing user base and a robust product-market fit. Holz's strategy underscores the value of building a product in close collaboration with its user community, utilizing their feedback for continuous improvement.

🤖 Harness AI for Disruption: Midjourney demonstrates the potential of AI to create groundbreaking solutions and disrupt markets. Embrace AI to innovate and differentiate your startup in competitive landscapes.

Background of Midjourney

Midjourney is one of the leading image-generative artificial intelligence companies in the world. It calls itself an "An independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species".

Within a mere two years since its inception, Midjourney has not only revolutionized the field of AI but also captured the public's imagination. This is evidenced by its prominent mentions, like John Green's engaging cat Tweet and a feature in a John Oliver segment, highlighting its widespread cultural impact.

As we step into 2024, a year anticipated to further elevate AI's prominence, let's start by spotlighting Midjourney and its remarkable journey in transforming both technology and imagination.

David Holz: The Genius Founder?

David Holz is the founder of Midjourney and he is a smart guy with experience in launching startups.

His journey began in Florida, where he ran a design business during his high school years, simultaneously delving into math and physics. Holz's academic pursuits led him to earn a Ph.D. in Applied Mathematics from the University of North Carolina. He further honed his research skills at the Max Planck Institute and NASA Langley Research Center, focusing on LiDAR, Mars missions, and atmospheric science.

Transitioning from academia, Holz founded Leap Motion, a venture capital-backed company pioneering in AR/VR technology. Leap Motion developed innovative hardware for hand motion tracking and device interaction, securing over $100 million in funding. Despite its initial promise, Leap Motion faced challenges in achieving a market fit and was eventually sold for $30 million.

After a 12-year tenure at Leap Motion, the demands of working in a high-stakes venture-backed environment weighed heavily on Holz, leading him to seek a new direction. It was during this period of reflection that he encountered a pivotal research paper from Cornell University on 'Denoising Diffusion Probabilistic Models'. With his deep technical understanding, Holz realized the potential of these models to revolutionize AI-generated imagery. This insight planted the seed for what would become Midjourney, marking the start of an exciting new chapter in his entrepreneurial journey.

Core Team

David Holz embarked on his new venture, Midjourney, in August 2021, after his tenure at Leap Motion. Starting with a core team of 10 skilled engineers, the company has maintained a lean team structure despite its rapid growth. Midjourney's hiring strategy emphasizes assembling a team with robust technical skills, and strong work ethics, and just trusting them to get on with it.

The effectiveness of Midjourney's culture is evident in its impressive financial performance. Boasting a staggering $5 million revenue per employee, the company outshines established SaaS giants like Dropbox, Slack, Squarespace, and Palantir. This remarkable figure not only underscores Midjourney's efficiency but also highlights its profitability, powered by a capable and dedicated team.

Midjourney's streamlined team size also allows for a flat organizational hierarchy. With minimal managerial layers, no board of directors, and small, autonomous teams, the company fosters an environment of independence and agility.

Additionally, Midjourney's unique incentive model offers employees profit-sharing instead of traditional stock options, aligning their interests directly with the company's success. This approach not only motivates the staff but also provides immediate financial benefits, setting Midjourney apart in both its operational structure and its approach to employee rewards.

Birth of Midjourney

With the team in place, the private demo of Midjourney was launched in September 2021. A few months later in Feb 2022, it was released to the public via Discord.

Midjourney's unique integration with Discord positions it in a dynamic social setting.

New users access Midjourney through Discord's server, where they can use text prompts, similar to ChatGPT, to generate images. However, Midjourney's platform adds a communal dimension. Users interact in various newbie channels, witnessing and drawing inspiration from each other's prompts and the resulting images. This collaborative environment fosters a sense of community and creativity.

Adding to its user-friendly interface, Midjourney introduces the ability for users to enhance their prompts with their images or a combination of multiple uploads. The AI processes these inputs, and within about a minute, presents four generated images for all the users in that channel. The user will then select the best image from these options.

Customer Acquisition

From the onset, David Holz identified Midjourney's primary user groups:

1. Advertisers: They use Midjourney to generate customizable images, offering an efficient alternative to traditional stock photos or hiring photographers.

2. Artists: These users leverage Midjourney for exploratory and conceptual work, rapidly visualizing ideas for clients before commencing actual projects.

To attract these key users, Midjourney recognized the need for virality, which is driven by a high volume of image generation.

Power of Discord

For Midjourney to quickly reach that stage, Discord helps in 3 ways:

• Community-Ready Platform:

  • Discord, popular among gamers, content creators, and streamers, aligns with Midjourney's target audience. This synergy allows Midjourney to harness the social nature of Discord, encouraging users to share their creations and organically grow the user base. As a result, Midjourney's marketing expenses are significantly reduced.

  • Given the controversial nature of AI-generated images, creating a place where all users of Midjourney gather also makes it easier when decisions have to be made. 

• Generating Volume:

  • The vibrant community on Discord actively creates and shares images. This visibility encourages more people to join, creating a strong viral loop.

    

  • In just a year, Discord's user base surged from two million to an impressive 14.8 million, reaching 18 million by 2023.

    

• Rapid User Feedback:

  • For an AI-centric product like Midjourney, user data, and feedback are vital for refining the model. On Discord, Midjourney engages users to choose from four generated images, aiding the AI in learning and enhancing future outputs. This quick feedback mechanism is instrumental in evolving the AI's capabilities.

Product Market Fit

At the heart of Midjourney's success is a decisive bias for action, a trait embodied by David Holz and his team. This proactive approach is complemented by the robust feedback loop from Discord, enabling continuous enhancement of Midjourney's offerings. While each release may not be flawless, this iterative process has rapidly evolved Midjourney to V5, with V6 already on the horizon, all within just 1.5 years.

Midjourney's strength lies in its vibrant community, which consistently generates high-quality images. This, coupled with the platform's ongoing improvements, has solidified Midjourney as a robust product in the market.

For new users, the journey to recognizing Midjourney's value is remarkably short. They simply have to:

1. Be a Discord member

2. Type in a prompt (or copy others in the group)

3. Get 4 amazing images

This ease of use and immediate gratification lead to quick 'Aha!' moments for users.

User-driven growth is central to Midjourney's strategy. The platform's users, predominantly artists, and advertisers, are the main drivers of its expansion. This organic, product-led growth model allows Midjourney to minimize marketing expenditures while effectively reaching and retaining its target audience.

This synergy of a strong product, engaged community, and efficient market penetration showed Midjourney's clear product-market fit.

The Business Model

Revenue

Midjourney's revenue model is primarily subscription-based, with plans ranging from $10 to $120 per month. Additionally, the company offers the option to rent extra GPU time, allowing users to extend their sessions seamlessly.

The higher tier plans allow users a private and faster experience. This contrasts with the public channel's bustling activity, where interactions from numerous users flood the screen. As of November 2023, Midjourney reported an impressive Annual Recurring Revenue (ARR) of $200 million.

Expenses

Their expenses are likely to fall into 3 categories: 

1. Data Collection: This involves gathering the necessary data to feed and train the AI models.

2. Data Cleaning and Training: These are critical steps for ensuring the quality and effectiveness of the AI, likely representing significant fixed costs.

3. Active Server Costs: These costs are variable, scaling directly with Midjourney's usage. They are tied to the number of prompts processed.

Among these expenses, the data collection cleaning and training of the model are likely to be fixed costs. In contrast, the active server costs fluctuate based on user activity. For Midjourney, and AI companies in general, the ability to secure GPUs and manage compute costs is crucial.

Efficient resource management directly impacts profitability and sustainability in the competitive AI landscape. Keeping server or compute costs low while handling increasing user demands is a key challenge that Midjourney must navigate to maintain its growth trajectory and market position.

Funding

Midjourney is completely bootstrapped with no outside money.

"We're self-funded. We have no investors. We're not really financially motivated. We're just sort of here to work on things we're passionate about and have fun. And we were working on a lot of different projects."

—David Holz

To be fair, such words can only be spoken by an entrepreneur that has money to spare and hence is "not financially motivated". But there are upsides for not raising a single cent from VCs:

1. Holz retains the power in decision-making

There are many unresolved issues with AI companies. Copyrights of the images used in the training for the bot, controlling the outputs of the bots, etc, the list goes on. From the view of the company, having a single voice with no outside influence might be critical for survival. No potential for a coup when opinions do not align (👀 OpenAI).

2. Unreliable VC capital

David Holz has voiced concerns regarding the sustainability of companies excessively dependent on venture capital. He points out the problematic cycle of burning through funds and perpetually seeking more investments, a scenario Midjourney has consciously steered clear of.

Additionally, Holz's skepticism extends to the reliability of venture capitalists. He observes that they invest in a hype cycle and exit once the bubble bursts. This perspective is likely colored by his own challenging experiences at Leap Motion, his previous startup. Despite raising over $100 million in venture capital, Leap Motion faced a disappointing outcome, selling for only $30 million in 2019.

Key resource

Holz understood from the beginning that what was needed wasn't money but computing power. By securing that, he can keep expenses low and also improve user experience. 

From the beginning, he leveraged his network and did many things to secure and lock down access to GPUs. Today, Midjourney actively searches for GPUs in parts of the world where it is nighttime to improve the time taken to generate images for users.

Projects such as the Andromeda Cluster founded by Daniel Gross and Nat Friedman which provides computing power to startups show an interesting shift from providing capital to compute.

Summing up what Midjourney is all about with this quote:

If you haven’t subscribed yet, join the other folks interested in starting your own company by subscribing here:

Subscribe now

Actionable Insights

This week's newsletter will be a different one. Instead of talking about tech startups, let's talk about a financial company that is renowned, Berkshire Hathaway. Here are key takeaways I have learned from the decisions that Berkshire Hathaway has made at the beginning of its journey.

🤝 Partnership is Key: The success story of Warren Buffett and Charlie Munger at Berkshire Hathaway highlights the strength of synergistic partnerships. Their blend of skills and perspectives can lead to outstanding business achievements. This partnership underscores the value of aligning with individuals who complement and enhance your strengths

🔄 Embrace Change: Berkshire Hathaway's transition from a faltering textile business to an investment giant showcases the importance of evolving business strategies. The company's shift to focus on high-quality companies at fair prices exemplifies the power of adaptability and strategic evolution in business

🌱 Prioritize Long-Term Value: Berkshire Hathaway's approach of focusing on long-term value rather than short-term gains offers a key lesson in business longevity

Background of Berkshire Hathaway

Berkshire Hathaway, shaped by Warren Buffett and Charlie Munger, symbolizes smart investing. Their strategy, focusing on long-term value, has consistently outperformed the market.

Each year, their "Woodstock for Capitalists", an annual meeting at Omaha, buzzes with wisdom, drawing crowds worldwide. This duo's simple, yet profound investment philosophy has inspired a dedicated following. In Berkshire Hathaway's story, Buffett and Munger have not just built a company; they've crafted a legacy of intelligent, ethical investing.

But it turns out that Berkshire Hathaway, as we know it today, is one of Warren Buffett's "dumbest" stock purchases ever. 

Berkshire Hathaway is a publicly traded textile company way that began way back in the 1800s. By the 1960s, it was struggling against competition and had lost all but 2 of its mills and 80% of its employees. Things were not looking good.

Maybe because of its gloomy prospects, its publicly traded stock was severely underperforming, so much that it was trading at a wide discount to the underlying assets that the company owned. 

In other words, the company is worth much more than the stock price is implying. 

Enter Warren Buffett

And that was the exact kind of company that Buffett cut his teeth on.

Warren Buffett's investment strategy in his early years was simple, to buy heavily discounted stocks with the hope of making one last sum of profits. In this aspect, Berkshire Hathaway was a real bargain.

By 1964, Buffett bought over a controlling interest in the company, and Berkshire Hathaway as we know it was born.

The company continued to be hammered by low-cost competition rising from not just the US but globally too. And there was nothing that Buffett could do to stop that. There was very little hope that the business would ever be consistently profitable. While the inventory was worth more than what Buffett paid for, it was the last puff of the cigarette with no real lasting value. In other words, a classic "cigar-butt" investment. 

Buffett “fought” the failing textile business for 20 years before finally giving up, calling this a massive failure that might have cost him a compounded return of $200 billion in opportunity cost.

The Pivot

Instead of selling the company and giving up, Buffett pivoted the Berkshire Hathaway business into his area of expertise, the stock market. Capital was reallocated from buying textiles to buying other cheap stocks in the stock market. The proceeds from those winners to buy even more. 

“I’d rather have a $10 million business making 15% than a $100 million business making 5%. I have other places I can put the money.”

—Warren Buffett

The cash-making Berkshire Hathaway machine is now slowly emerging from the ashes of a failed textile business.

Enter Charlie Munger

Berkshire Hathaway was on the brink of transformative growth when Charlie Munger, a long-time friend of Warren Buffett, joined the fold in 1978. This pivotal addition marked a significant shift in the company's trajectory.

The Meeting That Changed Everything

Munger was introduced to Buffett by a business friend (perhaps the most profitable introduction in history). In 1959, during a visit to Omaha, they first crossed paths and the chemistry between them was instant and undeniable. They soon found calling each other frequently to share investment insights and often ended up investing in the same companies.

Throughout the 1960s and '70s, their collaboration intensified. They became the largest shareholders in the Blue Chip Stamp Co., a shared investment venture that led to acquisitions such as See’s Candy, the Buffalo News, and Wesco.

Munger's Influence on Buffett: A Shift in Strategy

Munger was instrumental in changing Buffett’s investment philosophy from value to quality. 

Buffett, initially trained in Ben Graham's school of thought, focused on undervalued stocks. To buy heavily discounted stocks with the hope of making one last puff of profit. Berkshire Hathaway was an example.

Munger, however, encouraged him to seek out high-quality companies at fair prices. That is to buy wonderful companies at fair prices, rather than fair companies at wonderful prices.

This marked the end of Berkshire's "cigar-butt" investing style and a shift to a quality-based investing approach.

"Charlie shoved me in the direction of not just buying bargains, as Ben Graham had taught me. It was the power of Charlie's mind. He expanded my horizons."

—Warren Buffett

See’s Candies: A Testament to the Power of Brand and Pricing
The acquisition of See's Candies in 1972 marked a significant departure from Buffett's traditional investment strategy. Until then, influenced by his mentor Ben Graham, Buffett had focused on buying undervalued stocks, seeking short-term profits from distressed companies. However, See's Candies was different. The company was not a bargain in the traditional sense; it was priced higher than what Buffett would have typically considered a good deal.

Munger recognized the intrinsic value in See’s Candies that went beyond mere numbers. He saw the strength of its brand and its fiercely loyal customer base. This allowed for a strategic increase in prices over time without losing customers, thereby improving profitability. Munger's vision was that a strong brand could command customer loyalty and pricing power, key factors for long-term business success.

Buffett, though initially hesitant, was eventually convinced by Munger. The acquisition of See’s Candies turned out to be a masterstroke, yielding high returns consistently. It exemplified the shift from seeking short-term gains in undervalued stocks to investing in businesses with strong brand equity and the potential for long-term growth.

Berkshire Hathaway made more than 8,000% (80 times) since investing in the company, with its stake now worth over $2 billion from a $25 million investment in 1972.

Complementary Yet Similar

Munger's legal acumen and expertise in taxes and business operations complemented Buffett's financial prowess.

While Buffett was known for his desire to be liked and be in the spotlight, Munger was recognized for his wisdom and indifference to public opinion and the man in the background. Again, complementary traits.

Their shared humility and relentless pursuit of knowledge, along with being a soundboard for each other, further solidified their partnership. In their 56 years together, they never argued, a testament to their mutual respect and understanding.

Buffett himself acknowledged that Berkshire Hathaway's success was inextricably linked to Munger's inspiration, wisdom, and involvement. This partnership not only reshaped Berkshire Hathaway but also left an indelible mark on the world of investment.

Leaving you with some nugget of wisdom from Charlie Munger himself.

“I have a friend who says the first rule of fishing is to fish where the fish are. The second rule of fishing is to never forget the first rule. We’ve gotten good at fishing where the fish are,”

"All I want to know is where I'm going to die so that I never go there."

If you haven’t subscribed yet, join the other folks interested in starting your own company by subscribing here:

Subscribe now

Actionable Insights

Here are key takeaways I have learned from the decisions that Canva has made at the beginning of its journey.

🌟 Embrace Audacity in your Vision: Melanie Perkins and Cliff Obrecht, without a technical background, dared to challenge the status quo in design. Don't be afraid to pursue a bold vision, even if it contradicts the norm.

🎯 Start with a Niche Market: Canva's initial focus on yearbook design with Fusion Books exemplifies the importance of starting small. This means identifying a niche market where you can test and refine your idea. This approach allows for a deeper understanding of specific customer needs and provides a solid foundation for scaling your business.

🏄‍♀️ Unwavering Determination and Drive. Multiple setbacks and multiple closed doors did not faze Melanie. Despite that, she didn't wait for luck to find her; she actively sought it out and put herself on the path of opportunities, increasing the 'surface area' for luck.

Company Background

Canva is challenging the design world, proving you don't need complex software like Adobe Photoshop to create stunning visuals. It's democratizing design, making it accessible and effortless for everyone, not just professionals.

With Canva, sophisticated design tools become user-friendly. Whether crafting an eye-catching Instagram post, a polished presentation, or a stylish flyer, Canva streamlines the process. It's a creative ecosystem, offering a diverse range of templates for everything from business cards to YouTube thumbnails.

In the current landscape where content is king, Canva stands out as a vital tool. It enables users to create visually compelling content without the steep learning curve of traditional design software. Canva is redefining design, opening doors for all to explore their creativity.

Founder's Background & Vision

Melanie Perkins and Cliff Obrecht, the duo behind Canva, break the mold of typical tech startup founders. At the heart of Canva's origin story, then known as Fusion Books, is a pair of young visionaries, aged just 19 and 21. Neither Perkins nor Obrecht came from a technical background. Yet, what they lacked in technical prowess, they made up for with a deep understanding of design challenges. Perkins, a part-time design program instructor at a university, experienced the world of design firsthand, giving her a unique perspective on the industry's needs.

Despite not fitting the typical tech-founder profile, Perkins harbored a vision grander than most: to redefine the future of publishing. They were not deterred by their perceived disadvantages and they had the audacity to try it.

Birth of Fusion Books

Design tools such as InDesign and Photoshop were the dominant choice for designers. But they look like this (and they still do to this day).

Not very intuitive right? Being a design teacher, Perkins experienced this frustration personally and thought that there had got to be a better way to make the design simpler and more collaborative.

"It would take a whole semester for students just to learn the basics. I realised the future of design would be entirely different, that it would be online, simple, collaborative and accessible to everyone."

- Melaine Perkins, Co-founder Canva

With this vision, Perkins and Obrecht took their first steps toward revolutionizing design. So they started with a familiar domain that they as students knew, yearbooks.

Since neither have any web development skills, they hired a software development firm to create the first version of a website called Fusion Books. It is a website where it is easy for folks to design and collaborate on yearbooks. 

Once users are done designing, Fusion Books will handle the printing and delivery, monetizing the physical printing aspect of the process using their own printing presses.

With that, they are off the races.

Although they began in a niche field of high school yearbooks, Fusion Book was just a proof of concept, a litmus test for a much larger hypothesis: that design, in its current state, was a challenge for many.

If that is the case the technology they developed from Fusion Book is going to be much more powerful than just the yearbook market as we will see later.

Getting customers for Fusion Books

The customer acquisition process of Fusion Book was trying everything until something sticks. They experimented with:

1. Direct mail campaigns. They invested in paid advertisements, hoping to catch the eye of potential customers.

2. Attending expos. By being present where their target audience congregated, they increased visibility (Unfortunately there were more sellers than buyers).

3. Sending out samples. This hands-on approach allowed potential customers to see and feel the quality of their work.

According to Perkins, the most effective way was to send sample yearbooks and show their work to potential customers. 

Give me the Money and I will give you Canva

Fusion Book was bootstrapped. They had a $50K loan, a rented Xerox printer, and a $5k tax refund to use for marketing. However, this financial constraint was a great learning experience for Perkins and Obrecht.

• Allowed them to slowly figure things out

• Kept their voice which might have been drowned out by VC funding

• Required them to deeply understand their business and prioritize customer needs 

• Learned how to make profits, not revenue

Being bootstrapped is possibly the only route for the couple anyway. Their inexperience would mean it will be challenging to raise any sort of venture funding. 

Within 4 years of operations, Fusion Books has become the largest yearbook provider in Australia. But Perkins has a greater vision than just yearbooks. She had the whole publishing industry in sight and needed more money to achieve that vision.

Mel is definitely the grand visionary, she had the vision when she was teaching design at university. We just attacked the grand vision at a micro-scale.

- Cliff Obrecht. Co-founder Canva

Raising venture capital was a hurdle for Perkins. She was based in Australia and had no connections. Doors were closed at every turn. So she put herself out there. 

“But yeah, decided to give it a go because when you don’t have any connections, you don’t have any network, you just kind of have to wedge your foot in the door and wiggle it all the way through.”

- Perkins

Eventually, she found her big break in a serendipitous moment. At a business conference in Perth, Perkins met Silicon Valley investor Bill Tai. She got invited to kitesurfing (I know, stereotypical SV stuff) which opened doors for Perkins to meet more investors.

This meeting was pivotal, eventually leading to a $3 million seed round in 2013. And Canva was officially born.

Subscribe now

Canva Core Team

Being a non-technical duo, Canva needed a highly technical co-founder to join the team. The chance connection with Bill Tai again proved instrumental in this quest, providing them access to a network and talent pool previously out of reach.

Bill introduced them to Lars Rasmussen (Founder of Google Maps) who would help to connect engineers to the Canva team and also serve as a screener for potential engineers that Perkins found. 

This search for a technical co-founder was challenging and prolonged, largely due to Rasmussen's exceptionally high standards. 

"I tried absolutely every single thing under the sun to find an amazing engineer that would meet Lars’s bar that appeared to be inordinately high."

-Perkins

Persistence and determination eventually paid off. Perkins convinced Cameron Adams to join the Canva team. Adams, recommended by Rasmussen, brought a wealth of technical expertise and experience to the table. His background and skills perfectly complemented Perkins and Obrecht's vision, making him an ideal choice for the role of technical co-founder. 

"I was amazed by the beautiful way he’d coupled design and development. The little icons filled up magically, and the finishing touches on his product were absolutely stunning."

-Perkins referring to Adams

This hire is probably the most important hire for Canva in the early stages because it unlocked massive possibilities for the future of Canva and also signaled to investors that they are now a complete team. 

The Rise of Canva

With the funding and people in place, the first iteration of Canva was made. It was a breath of fresh air compared to legacy design tools. It allowed users access to:

• Pre-made Templates: Offering a variety of designs for various needs.

• Simple Design Guidelines: Making the design process more approachable.

• Rich Media Library: Over 1 million photographs, fonts, and graphics

All are available in a drag-and-drop style.

To further ease new users into the world of design, Canva developed a simple how-to guide.

This guide wasn't about complex design concepts; instead, it offered straightforward challenges like changing the color of a circle or adding a hat to a monkey. These small, engaging tasks were important in acclimatizing users to Canva's platform, reducing the fear factor often associated with design software.

Although the early version of Canva had templates and guidelines that were cheesy and restrictive, that was also kind of the point. It was aligned with Perkins's vision that design is supposed to be simple and easy.

Getting users for Canva

Armed with the experience gained from Fusion Books, Canva's founders approached customer acquisition for Canva with a refined strategy. They focused on two key areas: leveraging relevant communities and emphasizing Canva's practical use cases.

Leverage communities

Canva was to be released as a beta in 2013 before its official public debut. For its beta, it had 2 goals, to generate a buzz and to test its products. 

To generate buzz, Canva went where the design folks hang out. They reached out to the design community, to adjacent groups that needed some design help. On top of that Perkins and Obrecht also badgered the press to cover their seed group and reached out to many other channels like blogs, podcasts, and conferences to offer their audience early access.

Furthermore, users that are enthusiastic enough to tweet directly at Canva were often onboarded almost immediately and professional designers that emailed them could also skip the line. This allowed for a mix of initial beta users:

• Pro designers: Constantly suggest features for the team

• Prosumers: Test of the usability of the platform

Although this tactic might seem simple, the waitlist for the beta grew to almost 50k people before their public debut. 

Focusing on use cases

The second method that Canva used to acquire customers was to focus on the specific use cases of Canva. Basically, what would people use Canva for and make it extremely easy for them to do that? 

In 2013, Facebook was the cool thing and many small businesses were moving onto the platform and there was a clear need for designs that could capture attention on the platform.

And Canva provided that, with its easy-to-use templates, perfectly sized for Facebook posts. This simplicity was enhanced by features that allowed users to invite colleagues for collaboration and facilitated sharing with one-click buttons. 

It was a textbook example of Product-Led Growth, where the product's inherent qualities – its templates and collaboration tools – drove user acquisition and expansion. 

This strategic approach not only secured an initial user base for Canva but also laid the groundwork for its continued growth and popularity. By focusing on the product's utility and the needs of its users, Canva was able to establish itself as a go-to design tool, paving the way for its success in the years to follow.

Leaving you with a thought from Perkins herself.