Stripe Data Reveals an AI Startup Explosion: A New Business Paradigm Where New Company Creation Has Doubled

The Startup Explosion: A Wave More Powerful Than the Pandemic

In a deep conversation between Replit founder Amjad Masad and Stripe co-founder Patrick Collison, two core players in the tech world shared a stunning set of data: In March 2025, the rate of new business creation on Stripe's platform nearly doubled year-over-year — a surge that even exceeds the 50% year-over-year growth seen during the 2020 pandemic lockdowns.

Patrick Collison mentioned he had just had lunch with Stanford economist Nick Bloom. Bloom is a professor in Stanford University's economics department, renowned for his research on innovation, productivity, and management practices. He had authored papers arguing that per-capita innovation and productivity growth were declining — the so-called "Great Stagnation." This concept was first systematically articulated by economist Tyler Cowen in his 2011 book of the same name. The core argument: since the 1970s, total factor productivity (TFP) growth in advanced economies has slowed significantly. While technological progress has surged in the information sector, its penetration into the physical economy has been far less impactful than the eras of electricity and the internal combustion engine. Bloom and his collaborators further quantified this trend — they found that the number of researchers needed to sustain Moore's Law doubles every 13 years, meaning "research productivity" itself is declining. But now, both men agree: the Great Stagnation appears to be over. Productivity is re-accelerating for the first time in decades, and the massive explosion of entrepreneurship is one of the most underappreciated economic facts of our time. If AI has indeed broken this trend, the economic implications would be historic.

Stripe currently processes over 25% of Delaware corporate registrations, which means its data is highly representative. It's worth explaining that Delaware is the most popular state for company incorporation in the U.S. — more than two-thirds of Fortune 500 companies are registered there. This isn't because these companies operate in Delaware, but because the state has the most mature corporate law system in the country. Its Court of Chancery specializes in business disputes, with rich and predictable case law, and virtually all venture-backed startups choose to incorporate there as C-Corps. As a result, Stripe's data effectively covers a highly representative cross-section of the U.S. new business ecosystem. More critically, the pandemic-era startup peak never fell back to its previous baseline — even before the AI wave arrived, 2022–2023 had already maintained a new high. The arrival of AI sent that curve sharply upward once again.

Not a Bubble: Quality and Speed Improving in Tandem

Faced with such explosive growth, a natural question arises: are these new businesses just a flood of lightweight "vibe coding" projects? "Vibe Coding" is a term that emerged in 2024–2025 as AI coding assistants became widespread, first coined by former OpenAI researcher Andrej Karpathy. It describes an entirely new approach to programming: instead of writing code line by line, developers describe desired functionality in natural language to an AI, which generates the code. Developers just need to "feel the vibes" — roughly scan the output, accept what looks reasonable, and when errors occur, paste the error messages back to the AI for fixing. This approach dramatically lowers the technical barrier to software development, enabling non-programmers to build functional applications, but it has also raised concerns about code quality and maintainability.

However, Patrick gave a surprisingly reassuring answer: even amid this massive growth, the average revenue per business has actually increased slightly. In other words, not only has the quantity doubled, but quality hasn't declined.

Even more striking are the changes in speed metrics for AI startups:

• 20% of startups charge their first customer within 30 days, compared to just 8% in 2020
• Compared to the SaaS boom and marketplace boom eras, AI-era companies are reaching the $1 million, $10 million, and $100 million recurring revenue milestones in roughly half the time
• Breakout companies are achieving revenue growth at unprecedented rates

It's important to understand ARR (Annual Recurring Revenue), the core metric for SaaS businesses. It represents the total revenue a company can expect over the next 12 months based on existing subscription contracts. Unlike one-time sales revenue, ARR is predictable and has a compounding effect. In venture capital, ARR growth rate is a key driver of valuation: going from 0 to $1 million ARR proves product-market fit, $1 million to $10 million proves scalability, and breaking $100 million means entering IPO candidacy. AI-era companies reaching these milestones in half the time represents a qualitative shift in capital efficiency and market penetration speed.

A standout example mentioned in the conversation is Magic School — an educational AI product built by a teacher during the pandemic using Replit. It went from zero to $10 million ARR in just a few months and is now a company valued at $500 million. This case perfectly illustrates the new startup paradigm of "domain expert + AI tools."

Vertical SaaS: An Underestimated Trillion-Dollar Opportunity

When asked which technology area he's most bullish on, Patrick Collison's answer was unexpected: vertical SaaS.

Vertical SaaS refers to software solutions designed specifically for a particular industry or niche market, as opposed to horizontal SaaS — cross-industry general-purpose tools (like Salesforce's CRM or Slack's team collaboration). Typical characteristics of vertical SaaS include: deep integration with industry-specific workflows, compliance requirements, and terminology; lower customer acquisition costs (because industry communities are tight-knit); lower churn rates (because switching costs are high); but a relatively smaller total addressable market (TAM). In the past, the core challenge of vertical SaaS was the mismatch between development costs and market size — the engineering investment to build specialized software for ice rinks might not be recouped by a limited customer base. The emergence of AI coding tools has fundamentally changed this economics: when development costs drop by an order of magnitude, a vast number of previously "not worth building" vertical markets suddenly become profitable.

Patrick's logic is clear: the opportunities brought by software and the internet are far from reaching most industries. In many sectors, what still dominates are "1990s-style clunky systems" — terrible mobile experiences, zero AI support, and no collaboration features. AI has dramatically lowered the barrier to software development, enabling domain experts to build customized solutions directly for the industries they know best.

Several vivid vertical SaaS startup examples were mentioned in the conversation:

• A European entrepreneur developing management software for ice rinks, which is becoming a multi-million dollar business
• A yoga instructor in rural England who built a pop-up yoga class connection platform using Replit
• A "looks maxing" app that precisely captured a cultural trend among young people

Patrick also shared a practical framework for startup discovery: look for things that are popular among young people but have low social status. Cryptocurrency was exactly this a decade ago, and when Stripe was founded, "serving startups" itself was an overlooked market.

Global Entrepreneurship: Opportunities Beyond Silicon Valley

Regarding the geographic distribution of startups, Stripe's data reveals an interesting dual trend:

On one hand, startup activity is surging dramatically everywhere around the world. Whether it's Dubai, Thailand, Japan, or Germany, all are more conducive to starting a business now than they were 12 months ago. AI has lowered barriers to entry everywhere.

On the other hand, the truly explosive breakout companies are still more concentrated in Silicon Valley, and this concentration may even be intensifying. However, Patrick pointed out that this may be because many companies start elsewhere and migrate to Silicon Valley after gaining significant traction.

Amjad Masad's take on this is: Silicon Valley will increasingly become a "platform city," while vertical SaaS and niche solutions will decentralize, spreading into more niche communities. Products like ice rink software and rural yoga platforms are precisely the kind of things Silicon Valley would never think to build.

Will AI Labs Swallow Everything?

The most thought-provoking part of the conversation was the discussion about whether AI labs will monopolize all value. Patrick used an elegant analogy: without food there is no economy, but most economic value is not captured by food producers. AI will become extremely important and abundant, but that doesn't mean all value will be captured by AI labs.

Amjad added several key points:

1. LLMs are fundamentally an infrastructure technology, similar to computing itself — text in, text out, with a highly commoditizable interface
2. There's a Moore's Law-like progression curve: GPT-2 in 2019 could only be built by OpenAI; now ordinary developers can train similar models on a phone
3. Humanity's unique value lies in creativity and cultural insight — LLMs excel at executing known effective solutions but struggle to discover new things outside the distribution

On whether business moats will change due to AI, the two referenced Hamilton Helmer's 7 Powers framework. This is one of Silicon Valley's most influential business strategy frameworks, revered by executives at companies like Netflix and Spotify. Helmer's seven sources of durable competitive advantage are: scale economies, network effects, counter-positioning, switching costs, branding, cornered resources, and process power. He argues that lasting competitive advantage must simultaneously satisfy two conditions: creating differentiated value for customers (the benefit condition) and being difficult for competitors to replicate (the barrier condition).

Their consensus: fundamental business moats won't change because of AI, but the pace of competition will accelerate dramatically. This means that once a market inefficiency is discovered, competition will rush in quickly — which may be a challenge for Silicon Valley's large startups, but is actually good news for long-tail entrepreneurs, since those niche markets aren't worth fighting over for big companies.

From Idea to Business: Five Minutes Away

The conversation mentioned MedVee — a two-person company on track toward $1 billion ARR, built extensively using Replit. Its founder Matthew's working style is impressive: from idea to launching a project on Replit takes just five minutes. He explicitly stated he'd rather manage AI agents than human employees.

Amjad compared Replit's vision to Nintendo — a platform with end-to-end control of the experience, from ideation to deployment to operations to monetization, eliminating every friction point. Patrick used the philosophy of Lisp to understand this vision. Lisp is a programming language invented by John McCarthy in 1958, the second-oldest high-level programming language in history. Lisp's most revolutionary design concept is "homoiconicity" — code and data use the same representational structure, allowing programs to manipulate their own code just as they would manipulate data, enabling powerful metaprogramming capabilities. Code is data, and in Replit's world, the organization itself can become code. Patrick used this analogy to hint at a deeper transformation: when AI makes software development as simple as editing a document, an organization's business processes, operational rules, and even business models themselves can be "programmed" — dynamically adjusted, instantly deployed, continuously iterated, blurring the line between "running a business" and "writing a program."

Looking back at nearly 50 years of Silicon Valley history, every decade has brought fears about the death of startups — the Japanese threat in the '80s, Microsoft's monopoly in the '90s, the dot-com bust in the 2000s, and Google and Facebook's dominance in the 2010s. But the evidence shows that betting on entrepreneurship has always been a durable and rewarding wager. And now, whether viewed from short-term trends or a long-term perspective, this may be the best era for entrepreneurship in history.

Key Takeaways