OpenAI and Anthropic Are Racing to Become the Same Company. It's Already Public.

OpenAI is on track for annualized revenue above $40 billion, roughly double its run rate at the end of 2025, according to Bloomberg. A day later, Bloomberg reported that Anthropic told prospective investors its second-quarter revenue topped $11.5 billion, up more than fourteenfold from $787 million a year earlier and from $4.73 billion in the first quarter. Anthropic also posted positive adjusted operating income for the quarter, per documents the wire service reviewed.

Read side by side, the two disclosures look like proof that the companies have split into separate businesses. One has consumer scale, the other has enterprise margin. What each of them has done over the past six weeks says the opposite.

The numbers aren't the same kind of number

OpenAI released a run rate, which is an annualized projection off current performance. Anthropic released booked quarterly revenue, which is a figure an auditor can check. Bloomberg noted the two may not be calculated the same way.

The gap matters more than the caveat suggests. CFO Sarah Friar has said OpenAI exited 2025 above $20 billion annualized, and the company was near $24 billion in late March. President Greg Brockman told staff the run rate grew more than 20% month over month in July, per Quartz, which places much of the climb to $40 billion in the third quarter rather than the second. Work backward from those markers and OpenAI's booked second quarter lands somewhere near $7 billion to $8 billion, against Anthropic's $11.5 billion.

That is arithmetic off disclosed figures, not a reported number, and a public filing would settle it. But it points at something neither headline said: on trailing revenue, the company with roughly twenty times fewer users may currently be the larger one. CNBC reported that Anthropic booked about $10 billion across all of 2025, meaning last quarter alone exceeded its entire prior year.

Anthropic sells the expensive tokens

The most useful figure of the week came from neither company. Vercel's AI Gateway index, published August 11, found that Anthropic collected 65% of gateway spending on 30% of token volume, charging 4.4 times the average price of every other lab's tokens.

That explains the rest of Anthropic's numbers. It holds 54% of the enterprise coding market against OpenAI's 21%, and 40% of overall enterprise API spend against 27%, according to Menlo Ventures, an investor in the company. Its consumer app base sits near 56 million monthly users by Sensor Tower's estimate, a rounding error next to ChatGPT, and it has not mattered. Anthropic has won the tokens buyers cannot substitute at any price: long-horizon agentic coding, multi-file refactors in production repositories, work where a wrong answer costs more than an expensive one.

OpenAI runs the mirror image. Roughly a billion weekly ChatGPT users, a 33% adjusted gross margin after inference costs quadrupled during 2025, per Reuters, and monetization layered on top through subscriptions and a young advertising business.

Both are buying what the other has

Sort recent activity by direction and the pattern is hard to miss.

OpenAI is moving toward Anthropic's business. Bloomberg credits its revenue acceleration partly to coding software, with demand rising for Codex and ChatGPT Work. The company hired Dali Rajic, most recently president and COO of Wiz, as chief revenue officer, its second in under a year, to build out enterprise sales before the listing. In July it cut GPT-5.6 Luna by 80% and Terra by 20% while leaving flagship Sol untouched, per Benzinga, a pricing pattern that protects a premium tier rather than a volume one.

Anthropic is moving toward OpenAI's. It launched Opus 5 at half the price of Fable 5, its top model, and scrapped a planned September increase for Sonnet 5. It has been buying consumer brand advertising at a scale no enterprise API vendor would rationally buy. Each of those moves trades margin for reach.

Two companies with opposite strengths, each spending its advantage to buy the other's.

The models stopped separating them

The crossing became necessary once model quality stopped being a differentiator. GPT-5.6 shipped July 9 and Claude Opus 5 on July 24. Third-party trackers give Opus 5 the edge on SWE-bench Pro and agentic coding evals, while Sol leads on DeepSWE and Terminal-Bench. Artificial Analysis separates them by two points on a 61-point index.

The benchmark gap is also partly theater. OpenAI publishes SWE-bench Verified, Anthropic publishes SWE-bench Pro, the tests measure different things, and each lab reports the variant it wins. Once neither side can win on capability, the fight moves to business model, and there are only two worth having.

The company they're both becoming already trades

Follow the crossing to its endpoint and a specific profile emerges: consumer distribution at population scale, enterprise contracts with pricing power, an advertising surface, and owned silicon so the margin isn't rented from a landlord.

Alphabet reported Gemini above 950 million monthly active users in the second quarter, sells into enterprise through Cloud, runs the largest advertising business ever built, designs its own TPUs, and funds all of it from operations. It is also cutting Gemini Flash pricing into the exact tier its two rivals are fighting over. It appeared in none of this week's headlines.

Who can afford the trip

Each company's constraint inverts the other's, and neither crossing is fundable out of operations.

OpenAI has to buy margin. It carries roughly $600 billion in compute commitments through 2030 against $122 billion in committed capital, with projected cash burn near $25 billion this year and positive cash flow not expected until 2030, per CNBC. Enterprise revenue is the highest-margin thing available to it, which explains the sales hiring. Margin gets bought slowly, through procurement cycles and security review, and that burn schedule doesn't accommodate slow.

Anthropic has to buy reach, because the margin it already holds is the most perishable asset in the market. Vercel's index shows the four largest labs' share of token spend dropping below 93% for the first time in seven months, DeepSeek now second by volume at more than twice Google's, and average price paid per token down 13.6% in July. This publication has covered that compression before. Distribution ages better than a capability premium, which is why Anthropic is spending one to buy the other, and why it needs to move before the 4.4x closes.

Both companies filed confidentially within a week of each other in June. Bloomberg reported Anthropic wants the public market's funding capacity in order to hold its position. Neither disclosure this week is a victory lap. Both are financing events for a transition neither company can pay for privately.

Anthropic is expected to list first, as soon as this fall, working with Morgan Stanley, Goldman Sachs and JPMorgan. That sequencing sets the comparison every AI listing after it gets measured against. But the comparison itself is built on starting positions that are already moving, and the more useful question for anyone pricing either company is how far each one gets across before the money runs out. Both figures are preliminary, both companies are unaudited, and the destination they are spending hundreds of billions to reach has been occupied for a decade by a company that got there without the debt.