Stripe says new businesses on its platform nearly doubled year over year

Why it matters

Stripe's proprietary data suggests AI is accelerating company formation and early monetization, even as agents threaten thin software products. The opportunity is shifting toward founders with deep knowledge, differentiated markets and plans ambitious enough to justify years of work.

Patrick Collison (@patrickc), Stripe's cofounder and CEO, said the number of new businesses starting on the payments platform was approaching twice last year's level as of July 26th, the strongest annual jump Stripe has recorded.

Speaking at Y Combinator's Startup School, Collison said Stripe's internal data also showed improvement beyond company formation. The median new business was performing better than its year-earlier counterpart, he said, while a greater share of businesses were reaching revenue thresholds ranging from $1 million to $10 million. Companies formed through Stripe Atlas, Stripe's incorporation service, were also reaching their first revenue faster.

"I think the Stripe data would suggest there's never been a better time," Collison said.

The nearly 2x figure is a Stripe metric, rather than a count of new businesses across the economy. Collison did not provide the underlying number of companies or the period used to define a new business, and payment-platform signups do not measure how many companies survive. Stripe's position inside the startup economy still gives the result weight: Stripe reported in February that it directly or indirectly powered more than 5 million businesses and processed $1.9 trillion in 2025 volume.

That annual update offered an earlier view of the same acceleration. Stripe said its 2025 cohort grew about 50% faster than the 2024 cohort, while twice as many companies reached $10 million in annual recurring revenue within three months. Stripe Atlas formations rose 41% during 2025, and 20% of Atlas startups charged a customer within 30 days, up from 8% in 2020. Stripe also says Atlas now accounts for 25% of Delaware company formations.

Knowledge remains an advantage

Collison paired the growth figures with an argument against outsourcing too much thinking to AI. He compared accumulated knowledge to a processor's fastest memory layer: information already available in "cognitive L1 cache" can be retrieved and combined more quickly than a person can formulate a prompt and wait for an agent.

"For a long time to come, neuronal lookups will be much faster," Collison said.

His test is the hiring behavior of AI labs and companies such as Stripe, which continue to pay heavily for cognitive ability even as models improve. Collison said discarding the value of learned knowledge before evidence shows its benefits have been exhausted would be premature.

He applies the same standard to writing. Collison said he continues to write his own material because he dislikes model-generated prose and has yet to find an AI-written essay compelling. He also said he has never sent the prewritten responses that products including Gmail increasingly place in front of users.

That stance places a limit on one of the most common startup pitches of the AI cycle. Models can reduce the labor needed to research, code and operate a product. They do not remove the advantage held by a founder who already understands the market, technology or customer deeply enough to ask better questions and recognize a weak answer.

AI changes where startups should begin

Collison acknowledged that increasingly capable agents will eliminate some software categories. A general agent that can navigate interfaces, assemble workflows and execute tasks can absorb functionality that previously supported a standalone vertical application.

He rejected the broader conclusion that model providers will consume every startup opportunity. Large organizations struggle to pursue hundreds of priorities at once, even when they possess the capital, distribution and technical capacity to enter adjacent markets. Collison pointed to Google as a company that has succeeded across many fields without building every possible product.

"In general, I think that fear has been overstated," he said.

The competitive pressure may instead change how founders choose a starting market. Collison said the traditional lean-startup approach, in which a team finds a small opening through advertising or rapid tests and expands incrementally, becomes harder when AI lets many competitors identify and attack the same opening.

His alternative is greater differentiation at formation. Founders can use AI to start with a technically ambitious product or an unusual market position that would previously have required a much larger team. He cited frontier AI labs and defense technology company Anduril as companies whose initial plans did not resemble narrow, low-cost experiments.

The practical implication is that AI lowers the cost of attempting a large idea while raising the competitive pressure around obvious ones. A founder building a thin interface around a widely available model faces rapid replication from other startups, model providers and established software vendors. A company organized around proprietary knowledge, difficult operations or a market others have ignored has more room to compound.

Founders still have to want the company

Collison also warned prospective founders against treating company formation as the main decision. The harder question arrives if the startup works.

"What if you succeed?" he said. "And you raise money and you have customers and you have employees and a whole thing. Are you going to enjoy that? Are you going to want to work on that for 10 years, for 17 years, for 30 years?"

Collison has spent most of his adult life inside that outcome. Before Stripe, he and his brother John built Auctomatic, software for online sellers, and sold it in 2008. Patrick briefly attended MIT and later described Stripe to Congress as an effort to make accepting internet payments easier.

The brothers decided to pursue Stripe after attending Startup School in Berkeley in 2009. Collison recalled making the decision while walking back from dinner, reasoning that the project "probably won't be that hard." Stripe took almost two years to launch publicly in September 2011.

The return to Startup School gave Collison's formation data a convenient symmetry. Stripe offered attendees free Atlas incorporation, removing one more administrative cost for anyone who left the event ready to start. The larger bet is visible in Stripe's product strategy: more new companies create more demand for payments, billing, tax, fraud prevention and incorporation services. Collison's optimism about startup creation is grounded in Stripe's data, and Stripe directly benefits if that reading proves correct.