OpenAI at $1.4 Trillion: 35 Times Run Rate, 107 Times Recognised Revenue

OpenAI is reported to be seeking at least $30 billion at a $1.4 trillion pre-money valuation, with its public listing pushed to 2027. That price is 35 times the $40 billion run rate Bloomberg reported in August, and 107 times the $13.07 billion of revenue OpenAI recognised in 2025. The OpenAI valuation question is which of those two denominators you believe is the business.

Is OpenAI overvalued at $1.4 trillion? On recognised revenue, the price is 107 times 2025 sales, higher than any public software company of size. On the most recent reported run rate, it is about 21 times, close to where Anthropic raised in May. The answer depends on whether a run rate that grew 20% in a single month keeps compounding.

What $1.4 trillion is actually being measured against

Bloomberg reported on 29 September that OpenAI is targeting at least $30 billion in new funding at a valuation of around $1.4 trillion. The same report, relayed by Quartz, said the discussions are at an early stage and could change. Nothing is signed.

Two details matter more than the headline. The $1.4 trillion is pre-money, so a $30 billion raise would land the post-money at about $1.43 trillion. And Bloomberg had earlier reported OpenAI weighing a round at $1.2 trillion, so the ask rose by $200 billion between reports.

The anchor is the last closed round. OpenAI said on 31 March that it had closed $122 billion at a post-money valuation of $852 billion, anchored by SoftBank, Amazon and Nvidia. From there to $1.4 trillion pre-money is a 64% step-up in six months.

Chief executive Sam Altman has said a listing this year would be ill-advised, citing the company's safety work, and Bloomberg frames the round as runway while the listing is deferred. So this is a price set with a handful of large investors, not one tested by public order flow. The reasoning behind the delay is covered in the analysis of what a listing below $1 trillion would signal.

Run rate is not revenue, and for OpenAI the gap was 1.8 times

A run rate takes a recent period of sales and multiplies it out to a year. For a business doubling in months the two diverge badly, and the divergence always flatters the run rate.

How a run rate is actually built

The only published definition from either lab came out of Reuters Breakingviews. Karen Kwok, citing a person familiar with the matter, described Anthropic's method. Take the last 28 days of consumption-based sales and multiply by 13. Then multiply the monthly subscription take by 12, and add the two.

That is a 28-day window, so one strong month lifts the annual figure by the full multiple. OpenAI has not published its own definition, and Bloomberg noted in August that differences in methodology between the two companies make direct comparison complex.

What OpenAI's audited 2025 shows

Audited documents viewed by Ed Zitron and verified by the Financial Times show 2025 revenue of $13.07 billion, up from $3.7 billion in 2024. Total costs and expenses were $34 billion. Loss from operations was $20.92 billion.

The net loss attributable to the company was $38.53 billion. That includes a $41.55 billion loss on the fair value of convertible interests and warrant liabilities, tied to the for-profit conversion. It is an accounting charge, not cash, so the operating loss is the cleaner number to carry.

The Financial Times, reviewing the same documents, reported monthly revenue of nearly $2 billion by the end of 2025, and OpenAI itself said in March it was making $2 billion a month. Annualised, that is about $24 billion. So the exit run rate was roughly 1.8 times the revenue recognised across the year.

I treat that ratio as the honest growth number. For OpenAI in 2025, close to half of the exit pace had not yet been earned.

Only the first row is audited. The rest are company pace figures relayed by the press. Bloomberg's 29 September report put annualised revenue past $40 billion with 70% growth since July. On a $40 billion July base that is about $68 billion, which matches the Axios figure.

OpenAI's valuation multiple under four denominators

Put $1.4 trillion over each of those figures and you get four different companies. The spread between the highest and lowest multiple is a factor of 5, and all four are being quoted somewhere this week as "the" OpenAI multiple.

107 times on what was earned. No listed software company of any size trades at 107 times trailing revenue. It says investors are paying for a future the audited accounts do not yet contain, which every private AI round in 2026 does. The question is only how far ahead they are paying.

35 times on the August run rate. This is the headline multiple, and the one in this post's title, because it uses the last run rate Bloomberg reported with a stated comparison basis. Co-founder Greg Brockman told staff the monthly run rate grew more than 20% in July alone, which is the growth investors are extrapolating.

21 times on the late September figure. If the near-$70 billion figure is accurate, the round is priced at about 21 times run rate, close to where Anthropic raised in May. On that reading the price is not outlandish. It is simply unverified.

OpenAI vs Anthropic on the same basis

Anthropic is the only private comparison that matters, and it has published more. Its Series H announcement on 28 May stated a $65 billion raise at a $965 billion post-money valuation, and said run-rate revenue had crossed $47 billion earlier that month.

Anthropic at the Series H

$965 billion over $47 billion is 20.5 times, the cleanest private AI multiple on record because both numbers come from the same announcement. By the end of July the run rate had topped $65 billion, according to Reuters, up from about $9 billion at the end of 2025. On that figure the May valuation is 14.8 times.

Set the two labs side by side on the last Bloomberg-reported run rate for each. OpenAI's target is 35 times $40 billion. Anthropic's closed round was 14.8 times $65 billion. OpenAI is asking more than twice the multiple, on a run rate that was lower at the comparable reading. Anthropic's own listing plan is examined in the piece on its October Nasdaq plan and enterprise mix.

What a $2 trillion Anthropic listing would imply

TechCrunch, citing the Financial Times, reported that investors expect Anthropic to seek $2 trillion or more at its IPO, with 2026 revenue expected between $100 billion and $120 billion. $2 trillion over the $65 billion run rate is 31 times. Over the midpoint of that 2026 range it is about 18 times.

So the two converge once each uses its most optimistic denominator: OpenAI at about 21 times its latest figure, Anthropic at about 18 times its expected year. The difference is that Anthropic's number is a forecast investors have been shown, and OpenAI's is a pace reading from an unnamed source.

One caveat cuts the other way. Sacra has reported that Anthropic books revenue through cloud resellers on a gross basis, counting the end customer's full spend. Only an S-1 will settle it. If it is right, Anthropic's multiple is understated, and I would not treat 14.8 times as final until that document is public.

The public software comps, and the one that pays more

Private multiples only mean something next to prices that clear every day. Here are the five largest listed software names a public investor would hold instead.

Market values move daily; the date next to each is the one in the source. Palantir and Oracle are from The Motley Fool, 29 September. The ServiceNow and Palantir price-to-sales ratios are from a Fool comparison sourced to Financial Modeling Prep. Microsoft is from the Fool's largest-companies tracker as of 2 October. Salesforce revenue is from its fiscal 2026 10-K, and its and ServiceNow's market values from CompaniesMarketCap on 8 October. Multiples are our division, rounded.

Palantir is the only comp that pays more, and it is profitable

Palantir trades at roughly 71 times trailing sales, on $6.2 billion of revenue and about $3 billion of trailing profit. It shows what the public market demands for that price: guided growth above 80% and a profit margin near half of revenue.

OpenAI has the growth. It does not have the margin. In 2025 its cost of revenue alone was $7.5 billion against $13.07 billion of sales, before $19.18 billion of research and development. The economics of serving tokens at that scale are worked through in the piece on inference costs and AI margins.

Microsoft is the anchor the comparison should use

Microsoft is the honest benchmark, because it is both the largest listed software business and OpenAI's largest counterparty. It grew revenue 18% to $331.8 billion in fiscal 2026 and the market values that at 11.6 times. OpenAI paid Microsoft $17.2 billion in 2025 across research, cost of revenue and other lines, according to the audited documents. That is more than OpenAI's entire recognised revenue, and a share of it sits inside the denominator of Microsoft's 11.6 times. The money loop between the labs and their suppliers is mapped in the explainer on how the circular deals work.

Why the 2027 delay changes the denominator

A bridge round at a fixed price, followed by a year of compounding, is a bet on the denominator moving. The valuation is set today. The revenue eventually filed against it is whatever has been booked by the time an S-1 goes public.

Run the arithmetic, and treat it as arithmetic rather than forecast. If the run rate compounds at 10% a month from a $68 billion base, it passes $210 billion in 12 months and $1.4 trillion is under 7 times. At 5% a month it reaches about $122 billion and the multiple is about 11.5 times, roughly where Microsoft trades. At zero growth it stays at 21 times and the round has overpaid.

Nobody outside the company knows which path is live. July's 20% in a single month says the first path is not fantasy. The 2025 accounts say the recognised figure will trail the pace figure by a wide margin on any path. When an S-1 appears, read one ratio first: calendar 2026 recognised revenue against the September run rate. If it is near the 1.8 from 2025, the pace readings were honest. If it is far wider, they were built on windows that did not persist.

Where this argument is weakest

A post arguing the multiple is unverifiable should state where its own framing fails. Four places.

Trailing multiples on a business tripling are near-meaningless

Nobody pricing this round uses trailing revenue. On a forward model the multiples compress fast. Reuters reported that Anthropic is projecting 2028 revenue of roughly $190 billion to $200 billion. If OpenAI's investors hold a similar view, $1.4 trillion is about 7 times a 2028 number. The 107 times in the title is true and, to the people writing the cheque, irrelevant. I think the trailing number still matters, because it is the only one that cannot be revised.

Software companies are the wrong peer group

A SaaS business sells seats at 75% gross margin. OpenAI sells compute-backed output at a margin the 2025 accounts put near 43% before research spend. Its nearest listed analogue on the cost side is a cloud provider or a chipmaker, not Salesforce. Comparing to software flatters OpenAI on growth and punishes it on margin, and a cloud comparison would do the reverse.

The reports could simply be wrong

The $1.4 trillion, the $40 billion and the near-$70 billion all come from unnamed sources through Bloomberg and Axios. OpenAI has confirmed none of them and declined to comment on the leaked 2025 accounts. The round could close at $1.2 trillion, at a different size, or not at all. Every multiple here inherits that uncertainty.

A revenue multiple is a symptom, not a verdict. This post is about revenue multiples because no earnings multiple exists. A $20.92 billion operating loss has no price-to-earnings ratio, and that absence is the most important fact on the page. The broader comparison is in the dot-com versus AI bubble scorecard.

What the OpenAI multiple means for your own raise

You are not raising $30 billion. The multiple still reaches you, because every AI-adjacent term sheet in the next year will be anchored to it, in one of two directions.

If you are selling an AI product and raising, an investor will quote 35 times as the market and discount you from it. If you are buying, your vendor will quote the same number as proof of category momentum. In both cases ask which denominator is in use, then ask for the recognised figure next to it. A company that can only give you the run rate has told you something.

I would not sign a term sheet that prices my business on a run-rate multiple without the trailing recognised revenue stated in the same clause. How the market priced a smaller company on the same logic is in the Cursor valuation analysis, and the ranking of which AI revenue figures are real is in the revenue leaderboard.

If your own run rate is built on a 28-day window, know what moved in those 28 days. One enterprise pilot billed on consumption can lift an annualised figure by a multiple of its own size. That is not dishonest. It becomes dishonest when the window is chosen after the number is good, and the only defence is to fix the method in writing first.

Frequently asked questions

Where to start

Build one sheet with three columns and keep it open for the next year. Column one is every valuation figure you see quoted for an AI company. Column two is the denominator that was used, with the source and the date of the window. Column three is the most recent audited revenue for the same company, or the word "none".

After a month the sheet will show you which companies and which journalists are mixing the columns. That habit, more than any single multiple, is what protects you when the number quoted at you in a negotiation turns out to have been built on 28 good days.

Related reading