What AI Is Doing to Startups: The Data So Far.

The boom is real, but it predates the bots

New-business formation has been on a tear, and it's important to see that this started with the pandemic, not with ChatGPT. U.S. Census Bureau data shows monthly business applications climbing from under 90,000 a month two decades ago to well over 450,000 a month by 2025: more than a fourfold rise, at record highs. Analysis of state filings by Wolters Kluwer's CT Corporation puts new formations up around 26% since 2020, with roughly 5.5 million businesses formed in 2025. So the raw appetite to start something is at a generational peak, but that wave was rolling before LLMs arrived, and AI is riding it rather than causing it.

What has changed: the clock

The real shift is time compression, in both directions.

On the way up, AI-native companies are reaching scale at a pace that would have looked impossible five years ago. ICONIQ's 2025 software analysis found AI-native firms hitting $100M in annual recurring revenue in four to eight quarters, against eighteen to twenty quarters for even top-quartile traditional SaaS: two to three times faster.

On the way down, the same compression applies. CB Insights data suggests that of the 14,000-plus AI startups launched globally in 2024, something approaching 40% had shut within about two years — and these were funded teams with real traction, not hobby projects. The mechanism is the "wrapper trap": build a thin layer on top of someone else's model, and a single foundation-model update can erase your product overnight.

The lean-team revolution

If one statistic captures the era, it's revenue per employee. The old rule of thumb was that reaching $100M ARR took several hundred staff and five-plus years. AI-native companies are rewriting it: ICONIQ and venture analyses report firms crossing $100M ARR with fewer than 100 people. Cursor reportedly did it in about a year with roughly twenty; Lovable in eight months with around forty-five.

Where the money and the risk concentrated

Capital has stampeded toward AI. Of roughly $425B in global venture funding in 2025, AI startups took close to $210B — about half of everything. But the economics underneath are harsher than the headlines. AI application businesses run gross margins around 50–60%, well below the 70–90% that made classic SaaS so attractive, because inference and GPU costs eat into every transaction.

Most AI wrappers generate little or no revenue, and even the fast-growing AI-native firms under $100M ARR have been burning cash heavily; ICONIQ pegged their median free-cash-flow margin near −126%. Fast growth, thin moats, and expensive compute is a combustible mix.

The honest caveats

Two things keep this from being a tidy story. First, it is genuinely too early to separate the "AI effect" from the post-pandemic formation boom and the interest-rate cycle running underneath it. Second, the lean-team headlines are survivorship: for every Cursor there are thousands of startups that never found product-market fit at all. An MIT study found that some 95% of generative-AI projects never make it past the pilot stage.