Everyone talks about AI hype every day; almost nobody talks about the numbers. So I gathered the public figures — verified against financial press, official blogs and confirmed funding rounds — for 2026 so far. The short conclusion: the growth curves we're seeing have never existed before in the history of software.
Before we start, a note on method: I only use publicly sourced figures. When a number is a third-party estimate (rather than company-confirmed) I flag it as such. Even with that filter, what remains is hard to believe.
The curve that defines the era: Cursor
Anysphere (Cursor) was at one point the fastest SaaS in history to reach $100M ARR. That was January 2025. What happened next is the chart that defines this era:
From $100M to ~$4B in annualized revenue in 17 months, with around $2.6B coming from enterprise (B2B). Its November 2025 Series D valued it at $29.3B, and by April 2026 there were reports of talks to raise at ~$50B. For calibration: Salesforce took more than a decade to reach the revenue Cursor hit in three years.
The multipliers of 2026
Cursor is not an isolated case. This is the table for the year so far:
| Company | ~1 year ago | Now (2026) | Valuation |
|---|---|---|---|
| Cognition (Devin) | $37M ARR (May 25) | $492M ARR (May 26) — 13x | $26B ($1B round, May 26) |
| Lovable | $100M ARR (~Jul 25) | $400M ARR (Feb 26); ~$500M est. | $6.6B (Dec 25); reported talks at ~$12B |
| ElevenLabs | $3.3B val. (Jan 25) | ~$500M ARR est. | $11B (Series D, Feb 26) |
| Harvey (legal) | $100M ARR (Aug 25) | $300M+ ARR (Jun 26) | $11B (Mar 26) |
| Mercor | — | $1B ARR (early 26), profitable | $10B (Oct 25) |
| Replit | $150M annualized (Sep 25) | ~$525M est. (Apr 26) | $9B (Mar 26) |
And the backdrop: Anthropic went from a $9B run-rate at the end of 2025 to $47B in May 2026, and Claude Code alone hit $1B annualized within six months of launch. Demand for building with AI is the tide lifting all of these boats.
The most disruptive stat: revenue per employee
What truly breaks mental models is not how much they earn, but with how few people. Annual revenue per employee, in millions of dollars:
For perspective: a well-run traditional software company sits around $200,000–400,000 per employee. The cases above are 5 to 40 times higher:
- Gamma passed $100M ARR with about 50 people, profitable for over two years.
- Midjourney earns ~$500M with fewer than 110 employees and has never raised a dollar of venture capital. Ever.
- Mercor crossed $1B ARR with ~200 people and positive cash flow.
- Cursor operates ~$4B with around 300 employees (an estimate; the figure isn't official).
The small ones: solo founders who exited through the front door
Unicorns get the headlines, but the more replicable stories are elsewhere:
- Base44 (Maor Shlomo). A single bootstrapped founder, building with AI. Launched in early 2025, hit $1.5M ARR in 4 weeks and $3.5M with 300,000 users in 6 months. Wix bought it for $80M in cash six months after launch — and subsequent milestones paid him an extra $38M. Eight employees at the time of sale.
- Cal AI (Zach Yadegari and Henry Langmack). Two founders who started at 17: an app that counts calories from photos using AI. $30M in revenue in 2025, $5.7M in January 2026 alone, and in March — at $50M ARR — MyFitnessPal acquired it.
💡 The pattern they share: neither invented new technology. They applied models anyone can call over an API to a boring, concrete problem, and executed faster than everyone else. The edge is no longer access to AI — it's iteration speed on a use case.
And underneath it all: AI is already writing the code
These successes float on a measurable structural shift. Google has been publishing it with dates:
At Anthropic the reported internal figure exceeds 90%. And at Y Combinator — the best window into the immediate future: already in the W25 batch, 25% of startups had 95% of their code AI-generated; at the W26 Demo Day (March 2026), 14 startups arrived with over $1M ARR before Demo Day — triple the previous year and an all-time record — with average growth of 14% per week, the highest YC has recorded in 20 years.
The patterns that repeat
After looking at all these cases, the common denominators I see:
- Tiny teams by design, not for lack of money. Lovable raised hundreds of millions and stays at ~250 people. The headcount constraint is strategy: less coordination, more agents.
- Speed as the only moat. Almost none of them have defensible technology per se; they have a shipping cadence competitors can't match.
- Distribution first. Cal AI grew through social media before the product matured; Lovable turned every user into marketing with "look what I built."
- B2B pays for the curves. Cursor's and Devin's acceleration comes from enterprise contracts (Goldman Sachs, Citi and NASA are Cognition customers), not individual developers.
The fine print
To be fair to the data: several 2026 ARR figures (Lovable ~$500M, Replit ~$525M, Claude Code) are estimates from trackers like Sacra, not official confirmations; Lovable's round at ~$12B was still in negotiation as of this writing; and for every success story there are hundreds of GPT wrappers that died without a headline — survivorship bias in this list is maximal.
But even discounting all of that, the signal is clear: in 2026 there is no longer a correlation between team size and business size. And for anyone who knows how to build software, that is the best news in decades: it has never been this cheap to try.
