Anthropic at $965B, OpenAI at $852B: Why the Higher Mark Was the Cheaper Price

Anthropic's $965 billion Series H in May put a higher number on a younger revenue line than OpenAI's $852 billion round in March. That headline reads as investors overpaying for the challenger. It is the reverse. Measured against the run rate each company disclosed when its round priced, the Anthropic valuation was 20.5 times revenue and OpenAI's was 35.5 times.

Why is Anthropic valued higher than OpenAI? Anthropic's $965 billion round priced in May against a $47 billion run rate that had grown about 5 times since the end of 2025. OpenAI's $852 billion round priced in March against roughly $24 billion. Investors paid a lower multiple for faster growth and a lighter compute bill.

The Anthropic valuation is a higher mark at a lower multiple

The two marks were set nine weeks apart. OpenAI closed $122 billion at an $852 billion post-money valuation on 31 March 2026, the largest private round on record at the time. Anthropic followed on 28 May with $65 billion at $965 billion post-money.

The gap is $113 billion, about 13%. TechCrunch's coverage of the Series H described it as possibly Anthropic's last private raise before a listing. That makes the $965 billion the price setting the floor for an IPO, which is why the comparison matters beyond bragging rights.

A post-money valuation is the price of the whole company including the new cash just raised. It says nothing on its own. You only learn what investors believed by dividing it by something the company actually earns, and the choice of divisor decides the story.

My position is simple. The headline "Anthropic overtakes OpenAI" is the least informative way to read these two rounds. The informative reading is the ratio, and on the ratio Anthropic was the cheaper asset on the day its investors signed.

Anthropic vs OpenAI revenue on the day each round priced

The fair test uses what each company told its investors at the moment of pricing. Later revenue readings are hindsight, and neither round's investors had them.

When the round closed, OpenAI said it was generating $2 billion in revenue per month. Annualised, that is about $24 billion. It also cited more than 900 million weekly consumer users, over 50 million subscribers, and business customers at about 40% of revenue, according to TechCrunch.

$852 billion over $24 billion is 35.5 times. That is the multiple OpenAI's March investors accepted.

Anthropic's announcement said its run-rate revenue "crossed $47 billion earlier this month." $965 billion over $47 billion is 20.5 times. Reuters reported in August that the run rate stood at about $9 billion at the end of 2025, so the figure investors priced had grown roughly 5.2 times in five months.

That growth rate is the single most important number in this post. Investors paid 42% less per dollar of run rate for a business compounding several times faster. On those facts the Anthropic round was not aggressive. It was disciplined, and I would rather have held the May paper than the March paper.

Anthropic vs OpenAI: valuation, revenue, losses and compute in one table

Every cell below comes from a company statement, an audited figure relayed by the press, or a reported draft filing. Hover any figure for its source and date. Where a cell is our arithmetic, it says so.

Only OpenAI's 2025 figures come from audited documents, and those reached the public through a leak. Anthropic's 2025 and H1 2026 figures come from a draft S-1 described by Fortune and the Wall Street Journal. No Anthropic S-1 was on EDGAR when this was written.

Read the table by row, not by column. On run rate, Anthropic is larger and was larger at its own close. On booked 2025 revenue, OpenAI is nearly three times the size. On commitments, OpenAI has promised about three times as much against a smaller valuation. The interesting row for an investor is the last one, and I return to it below.

The gross versus net fight, and how much it moves the Anthropic multiple

The strongest objection to everything above is accounting. Bloomberg reported on 9 October that the Financial Times traced a gap in OpenAI revenue estimates to investors applying "the same methods as rival Anthropic PBC, which accounts for sales from cloud providers differently."

When a customer buys Claude through Amazon Bedrock or Google Vertex, Anthropic records the full customer spend as revenue and books the cloud platform's cut as an expense. TradingKey's reading of the prospectus materials says Anthropic argues it is the primary obligor because it sets prices and provides the service. A net presenter would record only what it keeps.

That choice inflates the top line and leaves profit unchanged. For a valuation built on a revenue multiple, it matters a great deal.

Reuters, as relayed by TradingKey and The Motley Fool, reported that Anthropic made about $2.16 billion through the Amazon and Google marketplaces in 2025, 47% of revenue, and paid about $351 million in fees. That is 7.6% of total 2025 revenue.

Strip the same share out of the $47 billion run rate and you get about $43.4 billion. $965 billion over that is 22.2 times. The adjustment costs Anthropic under 2 turns of multiple. It does not come close to closing the 15-turn gap with OpenAI.

The third row is a stress case, not a forecast. Marketplace share climbed from 11% to 47% in two years, so assuming 2025 holds may flatter Anthropic. Even doubling the fee load leaves it at about 24 times.

The fight cuts both ways. Bloomberg's account implies OpenAI's figure rises when measured the Anthropic way, since investors computing it that way reached roughly $70 billion against OpenAI's own $50 billion. If OpenAI also understates on a like-for-like basis, the gap between the two multiples narrows from the OpenAI side, not the Anthropic side. The broader problem of which AI revenue numbers deserve trust is ranked in the AI revenue leaderboard.

Losses at both labs are mostly non-cash, and that changes the comparison

Both companies report net losses that dwarf their revenue. Both losses are dominated by the same kind of charge, and reading them at face value misleads in the same direction.

Anthropic's first half of 2026

The Wall Street Journal, reviewing the draft prospectus, reported an $80.7 billion net loss on $16.3 billion of revenue for the first half. Of that, $72 billion came from non-cash adjustments on convertible notes and financing instruments. The residual is about $8.7 billion, our subtraction.

A rising valuation makes convertible instruments more expensive to carry at fair value. So the bigger Anthropic's mark, the bigger its reported loss. That is an accounting artefact of success, and the prior analysis of Anthropic's $42 billion 2025 loss shows the same pattern a year earlier.

OpenAI's 2025 on the same basis

OpenAI's audited 2025 accounts, as reported by Zitron, show the same structure. Net loss attributable was $38.53 billion, including a $41.55 billion fair-value loss on convertible interests and warrants. The cleaner figure is the operating loss of $20.92 billion, which is 160% of revenue.

Anthropic's 2025 operating loss was $8 billion-plus on $4.6 billion, roughly 174% of revenue on Fortune's figures. So in 2025 the two labs lost money at a similar rate per dollar of sales. The difference shows up in 2026, and it is the reason I think the market rewarded Anthropic.

The operating profit signal

Fortune's summary of the draft says Anthropic expects to be profitable on an operating basis for a second straight quarter. TechCrunch, citing the Wall Street Journal, reported in May that the company expected a 130% revenue surge to bring its first operating profit. OpenAI's own presentation, per the FT, projects negative free cash flow of $278 billion from 2026 through 2030.

Those are different kinds of documents, one a draft filing and one an investor deck. Still, one lab is describing operating profit now and the other is describing five more years of burn. A valuation premium for the first is not irrational.

Compute commitments against price: the row that explains the AI economy

Every frontier lab has signed multi-year compute contracts. These are mostly obligations the company must pay whether or not demand arrives. The Seoul Economic Daily's translated report of the draft described Anthropic's as mostly contracts that cannot be terminated early.

Sam Altman wrote in November 2025 that OpenAI was "looking at commitments of about $1.4 trillion over the next 8 years." Against the $852 billion mark, that is 1.64 times the company's price. Anthropic's draft reportedly lists about $450 billion of cloud agreements over a decade, 0.47 times its $965 billion mark.

Measured against the run rate each disclosed at its close, the gap widens. OpenAI's commitments were about 58 times its $24 billion March run rate. Anthropic's were about 9.6 times its $47 billion May run rate.

Why this row matters more than revenue

A committed compute contract is a fixed cost that revenue must grow into. The lower the commitment relative to revenue, the less growth a company needs before the contracts stop being a risk. That is why the commitment ratio, not the run rate, is the cleanest single signal of how much execution a valuation demands.

The ratio also connects these two rounds to the wider circular financing between labs and their suppliers. Amazon, Nvidia and Microsoft participated in OpenAI's round. Amazon contributed $5 billion to Anthropic's, per TechCrunch, and Anthropic's three largest reported cloud contracts are with Google, AWS and Microsoft. Each lab's investors are, in part, its vendors. How those obligations travel through lessors and lenders is traced in the piece on off-balance-sheet data centre leases.

What investors appear to be paying for in the Anthropic valuation

Put the rows together and a coherent picture emerges of what the May investors bought that the March investors did not. I see four things, in order of weight.

1. Velocity. A run rate up about 5.2 times in five months, from $9 billion to $47 billion, and to $65 billion by July. Investors pay for the second derivative, and Anthropic's was steeper.

2. Usage-based enterprise revenue. TradingKey reported that about $3.8 billion of Anthropic's 2025 revenue came from usage and $789 million from subscriptions. Usage revenue scales with customer workloads such as coding agents rather than with headcount. OpenAI's strength is consumer reach, 900 million weekly users, which is a different and slower monetisation path.

3. A lighter fixed-cost burden. The commitment ratio above. Less of Anthropic's future is already spent.

4. Proximity to operating profit. A reported quarter of operating profit is worth more than a projected one in 2030, even at private-market discount rates.

What investors were not paying for is equally clear: booked history. On 2025 revenue the Anthropic valuation is roughly 210 times. Nobody pays that for a trailing year. They pay it for a trajectory, and the remaining question is whether the trajectory survives contact with a public market. The March round's own logic is examined in the analysis of OpenAI's run-rate multiple, and Anthropic's listing path in the piece on its IPO and enterprise mix.

Where this comparison is weakest

A comparison built on leaked drafts and anonymous run rates should state its own failure points. There are four, and the first is serious.

The Anthropic figures are a draft, not a filing

Every 2025 and 2026 Anthropic number in the table comes from a draft prospectus reported by Fortune, Reuters and the Wall Street Journal. No public S-1 existed on EDGAR when this was written. The sources disagree on detail: customer concentration is about 20% from two customers in one account and a quarter in another. Commitments are $450 billion, $518 billion or up to $750 billion depending on the report.

Run rates on 28-day windows exaggerate momentum

Both multiples at close rest on company pace figures. A single strong month lifts an annualised figure by its full multiple. Anthropic's H1 2026 recognised revenue of $16.3 billion annualises to about $33 billion, well below the $47 billion run rate it announced in May. That is the honest correction to my headline: the gap between pace and booked revenue is wider at Anthropic, because it is growing faster.

The rounds are nine weeks apart in a market moving monthly

Comparing March to May treats them as simultaneous. They were not. AI private valuations rose quickly through spring 2026. Some of Anthropic's higher mark may reflect timing rather than relative merit, and I cannot separate the two with the data available.

The commitment figures are not like for like

Altman's $1.4 trillion is a 2025 statement of "commitments" over eight years, without a breakdown. Anthropic's $450 billion is a contracted cloud figure over a decade from a draft. They may not count the same things. I would treat the commitment ratio as directional until both companies file audited notes on purchase obligations.

Frequently asked questions

Where to start

If an AI vendor or investor quotes either lab's valuation at you this quarter, ask for the commitment ratio before the revenue multiple. It is the one number that tells you how much future growth is already pre-sold to a cloud provider, and it is harder to dress up than a 28-day run rate.

Then put the date next to every multiple you see. A $965 billion mark set in May against a May run rate is a fact. The same mark set against an October run rate is a flattering story told after the fact.

Related reading