Anthropic's $47B run rate overtakes OpenAI: what the leaderboard actually says

Update (2026-08-17): The article’s figures on OpenAI’s latest fundraise are now outdated. OpenAI has since closed a $110 billion round led by Amazon, Nvidia, and SoftBank, hitting a $730 billion pre-money valuation (with reports citing an $840 billion post-money valuation), according to TechStartups and Crunchbase. This supersedes the article’s reference to a “$122 billion round” at an “$852 billion post-money valuation,” and the new round is described as the largest venture deal ever, further underscoring the scale of the AI capital race.

Additionally, Anthropic has closed a separate $30 billion funding round at a $380 billion post-money valuation, per CNBC — the second-biggest private financing round on record for tech. This development is distinct from the article’s mention of Anthropic’s confidential IPO filing at a $965 billion valuation, and it adds a fresh data point on the company’s private-market standing ahead of any public listing.

The headline number on the AI Business revenue leaderboard is genuine news: Anthropic passed OpenAI in April, reaching roughly a $47 billion run rate by May 2026 against OpenAI’s estimated $24–33 billion, and filed confidentially for an IPO at a $965 billion valuation. The crossover happened months earlier than analysts predicted, and it flips a year that started with OpenAI comfortably ahead.

But the leaderboard’s own caveats matter more than the ranking. The two companies are not measuring the same thing. Anthropic reports revenue on a gross basis, counting total end-customer spend through cloud resellers like AWS, Google, and Microsoft as revenue and booking partner payouts as expense. OpenAI reports closer to net. That difference inflates Anthropic’s number relative to OpenAI’s, and it means the “challenger is ahead” framing needs a footnote before it becomes a verdict.

The real story is the growth rate, not the rank. Anthropic went from roughly $9 billion annualized at the end of 2025 to a ~$47 billion run rate by May 2026. That is more than a fivefold increase in about five months. OpenAI confirmed ~$2 billion in monthly revenue and closed a $122 billion round, the largest private raise in history, at an $852 billion post-money valuation. Both companies are compounding faster than any software business on record, and the leaderboard’s combined run rate across the ten listed companies lands near $87.5 billion, with the top two holding about 86% of it.

The business-model split explains the crossover better than any capability gap. Roughly 85% of Anthropic’s revenue is enterprise; roughly 85% of OpenAI’s is consumer subscriptions. Enterprise deals with cloud resellers book earlier and larger, and they flow through as gross revenue on Anthropic’s books. OpenAI’s consumer business is real but slower to annualize, and its net reporting makes its number look smaller next to Anthropic’s gross figure. The two are not apples-to-apples, and the leaderboard says so plainly.

What is genuinely new here is not the ranking itself but the correction buried in the middle of the page. The widely quoted “$2B ARR” for GitHub Copilot did not survive contact with analyst work. Microsoft has never disclosed Copilot’s ARR. What it reported in FY26 Q2 earnings is 4.7 million paid subscribers, up about 75% year over year. Independent analyst work built on that seat count and tier mix now puts ARR at roughly $0.9–1.1 billion, about half the figure most coverage carried. The leaderboard corrected its own row, and it deserves credit for doing so in public.

The same honesty applies to revenue per employee, the most abused number in AI coverage. The leaderboard previously carried the line everyone else carries: Cursor doing $2 billion with about fifty people, roughly $40 million per employee. That figure is wrong. Anysphere reported a team of over 300 in November 2025. Against a $2 billion run rate, that is about $6.7 million per employee. Still extraordinary by any pre-2023 software benchmark, still the highest evidenced figure on the board, but a sixth of the number in circulation.

Midjourney’s famous “$12.5M per head” has the same problem from the other direction. Public headcount estimates range from roughly 40 to 163 people, and revenue estimates run from Forbes’s ~$300 million for 2024 to the ~$500 million widely quoted since. Divide an estimate by a guess and you get a headline, not a fact. The leaderboard removed the calculation rather than pick the flattering pair, and that is the right call.

The insight survives without the arithmetic: AI-native companies earn multiples per head that traditional software never reached. The specific numbers do not survive a check of when the headcount was published. Most of what circulates online is guesswork dressed as precision, and the leaderboard now says so explicitly. That is a service to every builder who reads a per-employee figure and tries to benchmark against it.

The distribution of the board is a cliff, not a ladder. The median run rate is about $1 billion, and six of ten companies sit at or under that mark. A “top ten AI company by revenue” is, more often than not, a company doing hundreds of millions. The top two run ~$75.5 billion combined; the remaining eight share ~$12 billion. The model layer still owns the top of the board, even as the application layer scales underneath it. Cursor at $2 billion, ElevenLabs at $500 million, and Copilot at roughly $1 billion are real businesses, but they are an order of magnitude below the frontier labs.

Midjourney remains the efficiency legend, and its story cuts against the funding narrative. Somewhere between $300 million and $500 million in revenue on $0 raised, reported profitable by Forbes, with no outside funding and headcount not reliably published. Its financials are private, so nobody outside can verify the margin. The point stands without the arithmetic: you do not need billions to build a profitable AI company, though you do if you want to train frontier models and land on the top half of this table.

Profitability is the question the leaderboard quietly dodges, and its own table is honest about the answer: almost none of the largest companies disclose a profit. Anthropic projected its first operating profit of roughly $559 million for Q2 2026, but Q2 has closed with no public confirmation of the actual result. Midjourney is reported profitable with no audited figure public. OpenAI is widely reported to be lossmaking while scaling. Revenue and profit have come apart further in AI than in any software market before it, because training runs and inference are paid for up front while revenue arrives monthly.

The most profitable AI company is probably not on this list at all. The reliable money in this cycle has been made selling infrastructure, not models. Nvidia and the cloud providers are excluded from the leaderboard precisely because their AI revenue is not separable from hardware and infrastructure lines. That exclusion is the right methodological call, and it also explains why the board looks the way it does.

For builders, the takeaways are concrete. The market is real, and the ceiling on what a small team can build has moved. When the #4 company does $2 billion with a few hundred people, the opportunity at every level is visible. The cheapest on-ramp is still selling a service, and the premium goes to people who can wield AI, not just describe it.

For investors, the leaderboard frames the open question: two companies are compounding faster than any software businesses on record, and the AI IPO race is about to test whether public markets agree with the private multiples. Anthropic filed confidentially at $965 billion. OpenAI closed at $852 billion post-money. The run-rate growth is what drives the multiples, and the accounting difference between gross and net reporting will matter when those filings become public.

The direction is not in doubt. These are real revenues from real customers. The leaderboard’s own closing line is the right one: not a bubble, but a market reordering itself in real time. The next thing to watch is whether Anthropic confirms that Q2 operating profit, and whether OpenAI’s net reporting starts to close the gap when the gross-vs-net caveat is stripped away.