Top 3 by revenue
● 1. Anthropic $33.2B over nine months. The only filed number on this list.
● 2. OpenAI About $50B annualised, and not measured the same way.
● 3. Cursor Over $4B annualised, from a standing start in 2023.
● Read this first Only one of these ten figures is audited. The rest are claims.
Every AI revenue ranking you have seen puts these numbers in one column and sorts them. That column is not real. One of the ten below is filed with the SEC under penalty of perjury. Two come from company blogs. Two are chief executives posting on X. The rest are journalists quoting people who asked not to be named. Sorting them together produces a tidy table and a wrong answer, so this one shows you where each figure came from and what it counts.
The ranking
Figures are the latest publicly available for each company and cover different periods. They are not like-for-like. The next section explains why.
Why you cannot just sort this column
Anthropic counts partner sales. OpenAI mostly does not. Anthropic's filing shows cloud marketplace sales at 48 percent of revenue in H1 2026 - roughly half the business arrives through AWS and Google Cloud, and it is all in the number. Reuters has reported that OpenAI's figures exclude the full value of sales through its cloud partners. Strip an equivalent share out of Anthropic's and the gap narrows sharply. Neither company is being dishonest; they are answering different questions.
Nine months is not a year, and a run rate is not either. Anthropic's $33.2B covers January to September. OpenAI's $50B is an annualised run rate - one good month multiplied by twelve. Ranks three to ten are all run rates too. A run rate assumes the next eleven months look like the last one, which for a company growing at these speeds understates it, and for one that just closed a large annual contract overstates it.
Gross is not net. Mercor's figure is explicitly gross annualised run rate, before the splits paid out to the people doing the work. That is a real number and a different one from everybody else's.
Most of these are unaudited claims. Anthropic's S-1 is a legal filing. Cognition and Sierra published theirs. Two are chief executives posting on X. There is no payment verification behind any figure except the filed one.
The ten, in order
1. Anthropic (Claude) — $33.2B
Source: SEC amended S-1, Jan-Sep 2026. Q3 alone was $17.0B, which the filing describes as 86 times Q4 2024 revenue.
The only company here whose revenue has been through an SEC filing. H1 2026 was $16.3B including $11.6B in Q2, and the business leans heavily on cloud marketplaces at 48 percent of revenue. Reuters reported annualised revenue crossing $65B in July. Anthropic is offering 825 million shares at $70 to $78 on Nasdaq as ANT.
Best for: Enterprise deployment, long-context work, regulated industries ·
See the listing 2. OpenAI (ChatGPT) — ~$50B
Source: FT, confirmed by CNBC, Reuters and Bloomberg, 8 October 2026. An annualised run rate, reported net of most cloud partner sales.
The figure came in below the roughly $70B OpenAI had signalled earlier in the year, a gap of about $20B that drew more attention than the number itself. The accounting treatment is the thing to hold onto: counted the way Anthropic counts, OpenAI's figure would be higher than $50B.
Best for: Broadest consumer reach, the largest third-party ecosystem ·
See the listing 3. Cursor — $4B+
Source: Forbes, 8 June 2026. Annualised run rate for the AI code editor built by Anysphere.
The steepest climb on this list for a company founded in 2023. Cursor now sells inside The X Subscription following the SpaceX acquisition of Anysphere, which changed its pricing shape considerably - the entry Plus tier is Grok-only, and third-party models are charged at API list price on top.
Best for: Teams that want AI inside the editor rather than beside it ·
See the listing 4. Mercor — $2B+
Source: a CEO post on X, 6 July 2026. Gross annualised run rate, before payouts to contributors.
Mercor sells expert human data and evaluation work to the frontier labs. The gross figure makes it look larger than a net comparison would, because most of the money passes through to the experts doing the work. It is the clearest example on this list of why the measurement basis matters more than the digit count.
Best for: Labs buying expert evaluation and training data ·
See the listing 5. Higgsfield — ~$1.35B
Source: a CEO post on X, 10 November 2026. Annualised run rate, and the freshest figure in the top ten.
AI video generation, scaled to well over a billion annualised inside a market that barely existed two years ago. Being the newest number here also makes it the least corroborated - no filing, no journalist, one post.
6. Cognition (Devin) — $1B+
Source: the company blog, 25 September 2026. Annualised run rate, published by Cognition itself.
Devin crossing a billion matters because autonomous coding agents were the category most people expected to stall on reliability. Cognition also absorbed Windsurf, so windsurf.com now redirects to devin.ai and the two products share a pricing page - worth knowing if you are evaluating them as separate options.
7. fal — ~$800M
Source: Bloomberg via TechCrunch, 3 November 2026. Annualised revenue for the generative media inference platform.
fal is infrastructure rather than an app: it runs image, video and audio models for other people's products. Several companies elsewhere on this list are plausibly its customers, which is a reminder that these revenues are not all independent of each other.
Best for: Developers serving generative media models at scale ·
See the listing 8. Perplexity — $750M+
Source: The Information, 23 August 2026. Annualised revenue, and the oldest figure in the top eight.
An August number in a November article is a long time in this market, so treat it as a floor rather than a current reading. Perplexity is also one of the heaviest crawlers on the open web, which is how answer-engine traffic now reaches publishers.
9. Lovable — $600M+
Source: TechCrunch, 24 September 2026. Run rate, confirmed by a co-founder.
Prompt-to-app building, and the clearest evidence that non-developers paying to ship software is a real market rather than a demo. Pricing has moved more than once this year, so check the current tiers before budgeting.
10. ElevenLabs — ~$600M
Source: a TechCrunch CEO interview, 24 September 2026. ARR, from an interview rather than a filing.
Voice is the quietest large category in AI, and ElevenLabs has most of it. The number is roughly level with Lovable despite a far narrower product surface, which says something about how much people will pay for voice that does not sound synthetic.
What this list is actually useful for
Use it this way
● Picking a vendor to build on Revenue tells you survival odds, not product fit. Use it as a floor, not a shortlist.
● Judging a category Four of the ten are developer tools. That is where money is landing hardest.
● Reading a press release Ask three questions: what period, gross or net, and who published it.
● Comparing two companies Only valid if both count partner sales the same way. Usually they do not.
The practical read: if you are choosing a tool, revenue is a poor proxy for whether it fits your work, but a decent proxy for whether it will still exist in two years and whether support will answer. A company at nine figures has engineers to fix your bug. A company at six figures might ship faster and care more.
The category signal is sharper than the individual numbers. Four of the top ten are developer tools - Cursor, Cognition, Lovable and arguably fal. Two are frontier labs. One is search, one is voice, one is video, one is human data. If you are deciding where to spend a budget or build a product, that distribution is more informative than any single company's run rate.
FAQ
Which AI company makes the most revenue in 2026?
Anthropic, on the only filed figure available: $33.2B for January to September 2026, with Q3 alone at $17.0B, disclosed in its amended S-1. OpenAI is second at roughly $50B annualised, though the two are not counted the same way.
Is Anthropic really bigger than OpenAI?
On the published figures, yes, but the comparison is unsound. Anthropic includes cloud marketplace sales, which were 48 percent of its H1 2026 revenue. OpenAI's reported figures exclude the full value of partner sales. Counted identically the gap would be much smaller.
What is the difference between ARR and run rate?
In practice, little - both annualise a recent period. Neither is audited annual revenue. A run rate takes recent consumption and multiplies it out, so it moves sharply with one strong or weak month.
Which AI developer tool earns the most?
Cursor, at over $4B annualised as of June 2026, ahead of Cognition's $1B+ and Lovable's $600M+.
How reliable are these numbers?
One is a legal filing. Two are company blogs. Two are executives posting on X. The rest are journalists citing sources. None except the filing is independently verified.
Where does the AI revenue data come from?
Anthropic's figures are from its SEC filings. The rest are press reports, company posts and executive statements, each dated in the table above so you can judge how stale it is.