Anthropic’s annualized revenue run rate surpassed $65 billion at the end of July, outpacing OpenAI’s roughly $40 billion run rate in the same period, CNBC and Bloomberg reporting confirm. That sentence would have sounded like science fiction two years ago. Two AI labs, both still private, now generating revenue at the scale of mature Fortune 100 companies, and racing toward public markets at valuations that would have been unthinkable for software companies a decade ago.
But the headline gap is not the whole story. How the two companies count their revenue, who counts China, and what investors are paying per dollar of sales all complicate the picture. Here is what the numbers actually say.
The run rate race
Anthropic’s path to $65 billion is verifiable in the public record. The company raised $65 billion in a Series H funding round in May 2026 at a $965 billion post-money valuation, when its run-rate revenue was $47 billion. Preliminary second-quarter revenue exceeded $11.5 billion, up from $787 million a year earlier, and the company posted its first quarterly operating profit of about $559 million, as reported. The July update pushed the annualized pace to $65 billion, adding $18 billion in annualized revenue in about two months.
OpenAI’s trajectory converges on the same conclusion from a different starting point. The company ended 2025 with $13.07 billion in booked revenue and a $20.9 billion operating loss, the bulk of which was $17.2 billion paid to Microsoft for Azure compute costs. Revenue sat flat around a $25 billion run rate from February through June. Then it accelerated: Greg Brockman said revenue run rate jumped more than 30% month-over-month in July, and August brought it past $40 billion, roughly $15 billion of annualized revenue added in two months. OpenAI’s enterprise revenue has now passed its consumer revenue, CFO Sarah Friar recently told investors, a shift the company is using to argue it can win the same durable, multi-year contracts that drove Anthropic’s growth, TheStreet reported.
The old question, will AI ever make real money, has been answered. The new question is who keeps the money.
The accounting asterisk
Here is the nuance the newsletters flagged this week. Anthropic records the full value of partner sales and treats cloud providers’ cuts as expenses, while OpenAI records only its share on certain partner transactions. The difference is the principal-versus-agent distinction in accounting: when Anthropic sells through a cloud partner, it books the gross sale; OpenAI, in comparable deals, books only its net share. Axios reporting noted that reclassifying Anthropic’s sales to a net basis would cut the headline number by only about 6% to 10%, leaving a multi-billion-dollar lead intact. So the accounting explains part of the gap, not most of it. The rest is real: enterprise and API demand accelerated Anthropic’s monetization sharply in 2026 even as OpenAI kept the larger consumer footprint.
This matters because both companies have filed confidential IPO paperwork, and the SEC correspondence from those filings will eventually force the numbers into comparable shape. Anthropic could hit public markets as soon as this fall at a valuation of $2 trillion or more, which would make it the largest debut on record, the Financial Times reported. OpenAI, by contrast, will not go public in 2026: Sam Altman said last Saturday that AI safety concerns make this a poor time for a listing, while Anthropic could delay its own IPO until after the November U.S. midterm elections, according to tradersunion.
The China gap
Meanwhile, the revenue story has a second act on the other side of the Pacific, and it is brutal. Rhodium Group estimates, published September 17, found that all of China’s leading AI models combined generate only about 10% of the annual recurring revenue reported by OpenAI and Anthropic. ByteDance leads the Chinese pack at $4 billion in ARR, Alibaba at $2.4 billion, Z.ai at $1.8 billion, Moonshot at $1 billion, MiniMax at $800 million, and DeepSeek, the lab whose models spooked U.S. markets, at just $500 million.
Adoption is not the issue; Chinese models are everywhere. Monetization is. And the valuation math looks stretched: Rhodium estimated DeepSeek trades at about 163 times revenue and Moonshot at about 50 times, versus 34 times for OpenAI and 21 times for Anthropic. Financing constraints and equity market uncertainty could limit Chinese labs’ ability to scale, Rhodium warned.
The slowdown debate
Hovering over all of this is Anthropic CEO Dario Amodei’s September 12 essay calling on the global AI sector to slow the pace of capability improvements, warning of a future in which armies of AI agents overwhelm the internet and outstrip human control. OpenAI’s Altman and Elon Musk publicly backed the stance. Anthropic even launched Claude Opus 5.5 this week, per Morning Brew, while publicly debating whether to slow down, which is a remarkable posture for a company seeking a $2 trillion valuation.
What it means beyond Silicon Valley
You do not need to pick a side in the model wars for this to touch your money. If you hold an S&P 500 index fund, you already own the cloud giants and chipmakers whose fortunes rise with AI compute spending, and both Anthropic’s and OpenAI’s IPO filings will eventually land in the portfolios of the index funds and pension managers who buy new listings. The more immediate effect is in the job market: enterprise AI contracts are how these run rates get paid, which means the companies buying the software are reorganizing workflows around it. That is worth watching for anyone whose employer just signed a multi-year AI deal.
My take: the revenue race is real, the accounting asterisk is minor, and the China gap is the most underappreciated number in tech right now. A company doing $65 billion in annualized revenue does not need the hype cycle’s permission to exist. But when a lab worth a trillion dollars asks the industry to slow down while launching new models, investors should read that as confidence about its lead, not humility about its risks. The IPO filings, when they go public, will show which parts of these numbers survive daylight. Everything else is narrative.
Published September 24, 2026. Sources: TechCrunch, CNBC via TheStreet, Axios via TechDefused, Rhodium Group via CoinCentral, Reuters via tradersunion, Morning Brew.
