Stripe's New Business Signups Nearly Doubled Year-Over-Year—and They're Earning More Too
Source: Y Combinator | Published: 2026-07-31T17:29:06Z
Patrick Collison revealed at YC Startup School that new businesses registering on Stripe grew nearly 2x year-over-year, with median revenue among new cohorts up from last year and revenue milestone attainment rates rising across the board.
New Business Formation Is Nearly Doubling Year Over Year
The number of new businesses registering on Stripe this year is up nearly 2x compared to last year. Patrick Collison shared this figure at YC Startup School — calling it the largest single-year jump he's ever seen, including the 50% spike during the pandemic in 2020, which had previously stood as his benchmark.
What's more striking is the quality. You might assume vibe coding has just made it easier for anyone to throw together a website — more volume, less signal. But Collison says the median revenue of new businesses this year is actually higher than last year. However you slice it — $1M, $5M, $10M ARR — the probability of a new company hitting those thresholds is moving up. Companies incorporated through Stripe Atlas are also reaching their first dollar of revenue faster.
That led him to one of the boldest claims of the conversation: based on current trend lines, this may be the best time in history to start a company.
The L1 Cache for Knowledge
When asked how students should think about learning in the age of AI, Collison reached for a metaphor any programmer will recognize — Jeff Dean's famous latency numbers every engineer should know.
L1 cache access versus a network read spans orders of magnitude. Knowledge works the same way, he said. Sure, you can have an agent look it up or let a model handle it — but that's nothing like pulling it directly from memory. Neurons are still dramatically faster for retrieval for the foreseeable future.
His conclusion: abandoning deep learning before model capabilities plateau is a premature surrender. Stripe, the major AI labs, and other top companies are still paying premium salaries to recruit the most exceptionally sharp people — that's the clearest market signal available.
He Has Never Sent a Single AI-Drafted Message
Collison mentioned a personal habit: he still writes everything himself. Gmail and WhatsApp both surface AI-suggested replies. He hasn't sent a single one.
He doesn't dispute what the models can do — they can find counterexamples to conjectures like the Jacobian and outperform humans on certain dimensions. But he said he has yet to read a piece of LLM-written prose that he actually thought was good. His explanation carried a note of genuine puzzlement: maybe it's because "writing quality" is hard to optimize for with RL — the utility function itself is difficult to define.
For him, writing is both a form of human communication and a way of reasoning through a multidimensional reality. On that front, he thinks the models still have a clear gap.
A Sushi Dinner and the Seed of a $50B Company
Stripe began on an ordinary evening in 2009.
Patrick and his brother John were both still in college. They came to Berkeley for YC Startup School — back when YC was far less prominent than it is today and startups weren't yet a campus phenomenon. After the event, the brothers grabbed sushi in Potrero Hill, and on the walk back they talked through an idea that had been nagging at them: payments on the internet were absurdly hard. Endless forms, bank visits, bureaucratic friction, documents written in Latin.
Walking that street, they decided to build Stripe. In Collison's own words, the reasoning was simple — "we might as well, because it probably won't be that hard."
That was 17 years ago. He's still at it.
Dropping Out Twice — and an Urgency That Wasn't Necessary
Collison noted he holds an unusual distinction: he's dropped out of college twice.
First after freshman year, to co-found a startup with Harj. He returned to MIT for a year, then dropped out again to build Stripe. He grew up in Ireland and had barely thought about startups before coming to America. Silicon Valley infected him with a sense of what was possible.
Asked where that urgency came from, he was candid: part of it was the classic young-person "speedrun" mentality, and part of it was a belief that Silicon Valley's opportunity was a window that could close — if he didn't move now, the chance might be gone in three or four years.
Looking back, he thinks that was wrong. "Over the past several decades, Silicon Valley has kept producing opportunities" — that's the actual pattern. For students in the audience worried that they'll fall permanently behind if they don't drop out immediately, he pointed to a book called The Winged Gospel: when airplanes were first invented, people thought civilization was entering a new epoch, that the human species itself was being transformed. Aviation did change an enormous amount — but it didn't rewrite society from scratch.
He's skeptical of that now-or-never anxiety, and he doesn't think we're in the last few years where starting a company is still possible.
Ross Boucher and "Just-in-Time Development"
Stripe wrote its first lines of code in fall 2009 and had its first real production user by January 2010 — Ross Boucher, of a company called 280 North.
At that point Stripe could do almost nothing: accept card payments. Ross quickly surfaced reasonable requests. Can I see all my transactions? They built a small dashboard. Can I issue refunds? They added refunds. Then one day Ross asked: when do I actually get paid? They added payouts.
Collison calls this "just-in-time development" — not guessing ahead of time what users want, but responding to real user feedback continuously. Stripe stayed in private beta, adding users month by month, until its public launch in September 2011. Nearly two years.
That pace was unusual inside YC's "launch fast" culture, but Collison's logic held: if real users are continuously giving you feedback, you're not building in a vacuum — you're calibrating against reality. The timing of a public launch matters a lot less when that's already true.
An Empirical View on "Will the Big Players Eat You Alive"
Many students in the room shared the same anxiety: will AI labs just build your product directly once your category matures?
Collison broke it into two separate questions. The risk of model capabilities making certain vertical markets obsolete is real — it's already happening in some areas. But the broader question — whether large companies will sweep up every opportunity — he's far more skeptical of.
His reference point is Google twenty years ago. Every startup asked the same thing: what if Google builds this? Google had the money, the talent, the compute — it seemed nearly omnipotent. But even with all those resources, Google couldn't do everything. The inherent complexity of large human organizations makes simultaneously advancing 100 priorities far harder than it looks from the outside.
Stripe's own data points in a different direction: in the AI era, the future isn't consolidation — it's decentralization, with more winners.
The Limits of Lean Startup in the AI Era
Collison raised an interesting challenge to lean startup methodology. For the past two decades, "find a small wedge and iterate fast" has been the default playbook. But he thinks that approach has become far more vulnerable in the AI era, because the barrier to copying is dramatically lower.
What's notable is how many breakout companies started from unconventional premises — the AI labs themselves, Anduril, and a long list of successes over the past decade — most of which visibly defied lean startup logic from day one, placing large, wide-front bets upfront.
His read: the reason lean startup was the only option twenty years ago was capital scarcity and high organizational overhead. You literally couldn't afford big bets. AI has compressed the cost of building and organizing by an order of magnitude, opening the "go big from the start" path to far more founders. In an increasingly crowded market of incremental edge plays, radical differentiation may be the more rational strategy.