AI Revenue Rankings 2026: Anthropic Hits $65B, Few Turn a Profit

Two companies at the top, then a steep drop

The October 2026 rankings of AI companies by revenue show a market where two companies take almost everything and everyone else trails far behind. Anthropic told investors its annualized run rate passed $65 billion at the end of July. OpenAI's September run rate has been reported at about $50 billion by the Financial Times and Bloomberg, and at close to $70 billion by Axios.21 The next company on the list is Databricks, at about $7 billion. Cursor, which SpaceX now owns, was above $4 billion in June.21 After that, the next seven companies on the board sit between roughly $0.3 billion and $1 billion.21

For founders and the investors backing them, these rankings say more than who is winning. They show which business models are turning AI demand into money, and which ones are still spending capital to buy growth.

Anthropic's rise and the fight over how to count revenue

The year's biggest story is how fast Anthropic grew. Its run rate went from about $9 billion at the end of 2025 to $14 billion in February, more than $30 billion in April, $47 billion in May and more than $65 billion by late July.1 CNBC confirmed the July figure through three people familiar with an investor update and said it was about seven times the level of a year earlier.4 The company's own early-April disclosure said run-rate revenue had passed $30 billion, up from roughly $9 billion at year-end.2 That is the point where several outlets said Anthropic overtook OpenAI.35

The figures since then are just as large. Anthropic's Q3 2026 revenue was reported at $16.8 billion, up from $11.6 billion in Q2. SemiAnalysis estimated the company's Q3 run rate at about $74 billion, and investors at an October meeting projected $110 billion by year-end.9 Quarter-over-quarter growth has slowed to about 46%, though, so the company is still growing fast but no longer at its spring pace.9

The sources disagree on how to compare the two leaders, and readers should take that seriously. Anthropic books the full amount a customer pays through a cloud provider and records the provider's share as a cost. OpenAI counts only its own share of some partner sales.10 That accounting difference is part of why OpenAI's number ranges so widely. One analysis says the earlier reports of about $68–70 billion included partner revenue, while the cleaner figure is around $50 billion.9 The October leaderboard therefore declines to rank the two against each other. It does note that Anthropic leads on published quarterly figures, with more than $11.5 billion in Q2 against OpenAI's $6.7 billion.21

Anthropic probably earns more right now. Its quarterly booked revenue is a figure from actual results, not an annualized estimate, and on that measure it is well ahead. Still, gross versus net accounting makes the gap look bigger than a like-for-like comparison would.

Older rankings that are still online make this harder. One widely shared list still puts OpenAI at "$5B+" and Anthropic at "$2B+", numbers more than a year out of date.25 An August tracker showed OpenAI at $24–33 billion, which is a range rather than a single figure.22 In a market changing this fast, the date on a revenue number matters about as much as the number itself.

Enterprise versus consumer

The clearest business-model lesson in 2026 is that selling to companies has beaten selling to consumers, at least so far. Forbes reported that about 85% of Anthropic's revenue comes from enterprise and developer customers. OpenAI's mix is roughly the reverse: about 85% is tied to ChatGPT consumer subscriptions, and roughly 95% of those users pay nothing.6 A separate analysis says OpenAI's mix has since moved, with enterprise now the larger share.9

That difference shows up in profit. Anthropic reported positive adjusted operating income in Q2 2026 and told investors Q3 would also be positive.21 One analysis puts the Q3 adjusted operating margin at 11%, while cautioning that no GAAP figure has been disclosed.9 OpenAI, by contrast, loses money on purpose. Investor materials reported by the FT project about $278 billion of cumulative negative free cash flow from 2026 to 2030.21 Late-2025 plans put Anthropic's 2026 cash burn at about 33% of revenue, compared with 57% for OpenAI.9

Some caveats apply. Anthropic's profit figures exclude stock compensation and sit alongside very large compute spending.29 A draft S-1 reportedly lists $518 billion of cloud and infrastructure obligations in the coming years.9 Bloomberg described the Q2 figures as preliminary and said they could be revised once audited numbers appear in a prospectus.1 A profitable quarter does not prove the business will stay profitable over time.

The application layer: where startups can still win

For anyone looking for AI startup ideas, the leaderboard's most useful lesson is that the fastest-growing companies below the top two do not train frontier models. Cursor, Cognition and Lovable sell tools built on the labs' models, and each reached its current revenue range within about a year.16

Cursor shows how this works. Anysphere went from $100 million in annualized revenue in January 2025 to $1 billion in November 2025, $2 billion by February 2026 and about $4 billion by June.11 Roughly $2.6 billion of that, about 65%, came from enterprise contracts.11 Its pricing combines subscriptions from $20 to $200 a month with per-seat enterprise deals, and it adds a markup of about 20% on underlying model costs while steering usage toward its own cheaper in-house model.17 SpaceX bought the company in an all-stock deal valued at $60 billion, which closed on August 14.16 That price is about 15 times the company's revenue at the time.11

Cursor's early go-to-market approach was unusual. It reportedly reached $100 million in ARR without any marketing spend.18 Its growth came from individual developers adopting the product first, with large enterprise contracts following later, rather than from a traditional sales team.

The risk is that the labs Cursor depends on also compete with it. One market analysis estimated Anthropic's Claude Code at about $8 billion in ARR in May 2026, double Cursor's figure.13 An application company pays its supplier for every model call while that supplier sells a rival product. That is the main tension in building on top of the big labs.

Other application companies are growing quickly too. Cognition, which makes Devin and Windsurf, reported a run rate of nearly $900 million in September.21 That was up from $492 million in May, an increase of about 83%. Lovable's co-founder said it had reached $600 million in annualized run-rate revenue.21 ElevenLabs passed $500 million in ARR during the first four months of 2026.21 Glean went from $200 million in ARR in December 2025 to $300 million in May 2026.

Who actually makes money

The profitability data supports a lesson for startup operators: small, lean teams tend to stay profitable, and large headcounts tend not to. One analysis describes ElevenLabs, Gamma and Midjourney as profitable with very strong evidence.29 OpenArt reportedly reached about $70 million in ARR with around 20 employees.29 Harvey, the legal-AI company, is the opposite case: about $350 million in ARR and roughly 1,200 employees, and its cofounder says it is still unprofitable.29

Midjourney is the clearest example. It is reported to be profitable21 and has never disclosed a priced funding round.21 Its revenue was about $300 million in fiscal 2024.21 It shows that an application company can earn money efficiently without the capital needs of a frontier lab.22

Databricks is the most solid company in the middle of the table. It reported a revenue run rate above $7 billion, growth of more than 80% year over year in Q2, and positive adjusted free cash flow over the prior 12 months. It raised $5 billion at a $190 billion valuation in August. Its edge comes from years of enterprise data contracts that now carry AI workloads.22

Some headline figures need adjusting. Mercor is free-cash-flow positive, but about 60% to 70% of its gross revenue reportedly goes to contractors, so its $2 billion-plus annualized figure overstates what the company keeps.29 At xAI, estimated run-rate revenue of about $208 million sits against an estimated $37 billion of cumulative spending.28 One tracker puts the industry-wide gap to break-even, across labs and hyperscalers, at more than $900 billion.28

What it means

The 2026 leaderboard leads to three conclusions. First, selling to businesses and developers has proved more profitable than consumer-first AI, and Anthropic's profitable quarters are the strongest evidence so far. Second, the most common route to billions in revenue for a startup has been the application layer: a narrow use case, rapid self-serve adoption, then expansion into enterprise accounts. Third, revenue figures need careful reading. Run rates, gross bookings and net revenue are different measures, and lists that mix them can mislead.

The next test is public disclosure. Reports say Anthropic is preparing an IPO with a valuation target of $2 trillion or more.1 If that happens, its prospectus would give the first audited view of whether a frontier AI lab is a sustainable business, and outside estimates would no longer be the main source.