OpenAI’s $20 Billion Gap Was an Accounting Comparison

OpenAI told investors its annualized revenue reached about $50 billion at the end of September, some $20 billion below the figure the market had been trading. Chip stocks sold off on Thursday before people close to a new funding round said the company still expects to hit or exceed $70 billion by year-end.

The ChatGPT maker is in early talks to raise about $30 billion at a $1.4 trillion value. OpenAI declined to comment.

The $70 Billion Number OpenAI Never Printed

The $50 billion figure came from an investor presentation covering sales through September. A person familiar with the slides said the company also logged 77% run-rate growth in the third quarter and 107% in the enterprise book, which overtook consumer.

The higher number that ricocheted through markets last month was not a printed OpenAI forecast of current pace. Backers had been trying to line the company up with Anthropic, which books a wider top line, and they had been working from a roughly $40 billion August base. When OpenAI later said revenue had grown more than 70%, that higher starting point helped produce reports of about $70 billion.

A second path to the same headline was simpler still: take OpenAI’s net sales and add the partner-channel gross that Anthropic includes, so the two labs look comparable on a slide. Either way, the $70 billion was assembled for a comparison. It was not cash that went missing between August and September.

HOW THE CIRCULATED FIGURE WAS BUILT

• The Anthropic overlay: Investors grossed OpenAI’s partner-channel sales up so the run rate could sit next to Claude’s, which includes cloud-marketplace spend.
• The August base: A roughly $40 billion pace was treated as the jumping-off point, then “more than 70%” growth was applied to it.
• The product: Those two steps, not a cut in ChatGPT bills, are what fed reports of about $70 billion last month.

That arithmetic still leaves a fast business. Chief financial officer Sarah Friar wrote in January that OpenAI had more than $20 billion of run rate in 2025, up from $6 billion in 2024 and $2 billion in 2023, tracking compute that reached about 1.9 gigawatts. A September pace near $50 billion is more than double that 2025 mark in under a year. The growth rate that produced the grossed-up $70 billion on one base is the same rate that takes a net figure from the $30 billion range to $50 billion. The curve did not break. The footnote did.

Anthropic Counts Cloud Partner Sales in Full

Anthropic’s annualized revenue hit $65 billion by the end of July. On a raw slide, that now sits above OpenAI’s $50 billion. The ranking is an artifact of the ledger, because the two companies do not put the same dollars on the top line.

Anthropic counts cloud reseller sales on a gross basis, taking total end-customer spend through Amazon Web Services, Google Cloud and Microsoft as revenue and booking the platforms’ cut as an expense. OpenAI counts only the share that comes back to it, including from Microsoft. OpenAI has told investors and staff that Anthropic’s method swells reported revenue by billions of dollars. Anthropic has said it follows standard practice because it is the principal in the sale: it sets price, delivers Claude, and treats the marketplace fee as a cost.

A confidential IPO prospectus reviewed in late September shows how large that channel already is. Anthropic routed 47% of 2025 sales, about $2.16 billion, through Amazon and Google. It paid those platforms roughly $351 million in distribution fees, about 16 cents on each dollar of those sales, and booked the fees in sales, marketing and partnerships. Two unnamed customers each supplied 12% of 2025 revenue, and many large accounts can cut spend without a matching cut in Anthropic’s compute bills.

HOW THE TWO LABS COUNT A DOLLAR

U.S. principal versus agent revenue rules turn on who controls the service before it reaches the customer. A principal books the gross amount; an agent books a fee. Both treatments can be clean accounting. They cannot be stacked as if they were the same run rate, which is what Thursday’s tape did in an afternoon.

Chip Stocks Ate a Private-Company Footnote

OpenAI is still private, so the correction had nowhere to land except the public names that sell it chips, cloud and memory. The Nasdaq Composite fell 1.25% on Thursday, its worst session since mid-August, and the Philadelphia Semiconductor Index dropped 3.4%.

THURSDAY MOVES IN THE AI SUPPLY CHAIN

Those are the vendors whose orders are supposed to prove that model demand is real. Oracle is tied to a huge OpenAI compute contract. Nvidia, AMD and Broadcom supply the processors. Micron sells the high-bandwidth memory those racks require. CoreWeave rents the capacity. None of them books ChatGPT subscriptions, and all of them moved as if OpenAI had lost $20 billion of sales rather than restated a comparable.

The first public reading treated the gap as a bait-and-switch. A slower reading is harder on the buyers of those stocks: they have been using private-lab run rates as a daily index of AI demand, and they do not agree on what the index includes. When the net number replaced the grossed-up one, the index fell even though billed business, on OpenAI’s own slides, was still growing 77% in the quarter.

A $1.4 Trillion Price on $50 Billion

The presentation landed in the middle of a raise. OpenAI is in early-stage talks for about $30 billion of new capital at roughly $1.4 trillion before the money comes in, a round driven by investor demand with no term sheet locked. In March it closed a $122 billion round at $852 billion.

THE MULTIPLE THE NEW MONEY IS BEING ASKED TO PAY

• On $50 billion: $1.4 trillion is about 28 times the September pace.
• On $70 billion: the same asking price is about 20 times, a multiple the market had been using when the grossed-up figure still circulated.
• Cash on hand: Friar said in early October that the company is still “very well capitalized” after the March round.
• Listing clock: OpenAI filed a confidential prospectus in June and has pointed to a 2027 debut, not 2026.

Sam Altman, the chief executive, said in September that “right now would be an ill-advised moment to go public,” pointing to safety work after models acted in ways the company did not intend. The delay leaves the private raise as the only live price, and that price now sits on a smaller denominator than the one backers had been quoting to each other.

right now would be an ill-advised moment to go public

Sam Altman, chief executive of OpenAI, in September

A 28-times sales ask is still a growth price, not a distressed one. It is also a price that moves several turns when the run-rate definition changes, which is why a private footnote can take 5.5% off Oracle in a session.

Claude Goes Public on Gross Numbers

Anthropic is preparing to list as soon as November and has been meeting investors on a valuation near $2 trillion. It closed a $65 billion Series H in May at $965 billion, a different $65 billion from the July run rate, and filed confidentially on June 1. Independent research house New Constructs called the offering the “most ridiculous IPO of 2026” and put the firm at $150 billion.

The prospectus that leaked in late September is the first public look at what gross accounting buys and costs. Revenue rose 12-fold in 2025 to about $4.6 billion, while the operating loss topped $8 billion and the net loss reached $42 billion, the latter swollen by fair-value charges. Consumption fees were about $3.8 billion; subscriptions were $789 million. Compute commitments run to about $518 billion over the next decade, with roughly 80% noncancelable or payable even if usage falls short, including $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft.

market penetration at a scale we believe would be difficult for any single organization to directly replicate

Anthropic, confidential IPO prospectus, on Amazon, Google and Microsoft distribution

The same filing warns that those partners are also investors, suppliers and rivals, a mix that “creates complex dynamics that could give rise to conflicts of interest.” If Anthropic lists on gross revenue, the first audited AI comparable will be the wider definition. OpenAI’s net $50 billion will look small next to it until someone publishes a bridge, which is the comparison Thursday tried to improvise and then punished.

SoftBank Asks the Gulf for Another $100 Billion

The other balance sheet in the room is SoftBank’s. Masayoshi Son is seeking as much as $100 billion from Gulf sovereigns to scale bets that already include about $65 billion in OpenAI. Last month SoftBank sold $11.1 billion of high-yield bonds, the largest such corporate sale on record, with yields as high as 9.75%.

Gulf money is already inside the new OpenAI round. Abu Dhabi’s MGX has co-led prior financing, and UAE funds have been discussed as putting in as much as $10 billion of the $30 billion now on the table. The same investors are being asked to fund SoftBank so SoftBank can keep funding OpenAI so OpenAI can keep buying compute from Oracle and Nvidia. Thursday did not break that circle. It showed how fast the public end of it marks when the private end changes a definition.

THE MONEY TRAIL INTO THIS WEEK

1. March 2026: OpenAI closes a $122 billion round at $852 billion.
2. May 2026: Anthropic raises $65 billion at $965 billion.
3. June 2026: Both firms file confidential IPO papers; OpenAI’s listing talk shifts toward 2027.
4. End of July 2026: Anthropic’s run rate hits $65 billion on gross cloud sales.
5. September 2026: About $70 billion circulates as an OpenAI comparable; Altman calls a 2026 listing ill-advised.
6. October 8, 2026: OpenAI’s own slides show about $50 billion; Nvidia, Oracle and Micron fall.
7. October 8-9, 2026: People familiar with the raise say the year-end target is still $70 billion or more.

Son’s Gulf ask sits on that last line. If the September pace is the number that sticks, the $1.4 trillion value is 28 times sales. If the year-end target is the number that sticks, the multiple compresses, and the bond and sovereign money look cheaper. Thursday priced the first. Friday’s backers are selling the second.

The Year-End Target Is Still $70 Billion

After the close on Thursday, people familiar with the fundraising talks said OpenAI expects to reach or exceed $70 billion of annualized revenue by the end of 2026, led by enterprise. The company shared that outlook with investors as part of the $30 billion round. Getting from a $50 billion September pace to $70 billion by December is about 40% more run rate in one quarter.

Nasdaq 100 futures were about 0.8% higher on Friday as that target moved through the same desks that had just sold the suppliers. The bounce does not settle the definition fight. It shows that the number the market panicked over is still the number the raise is built on.

WHAT WE KNOW

• September pace: OpenAI told investors annualized revenue was about $50 billion at month-end, with 77% quarterly run-rate growth and 107% in enterprise.
• The gap: About $20 billion separates that net figure from the roughly $70 billion that had been circulating as a comparable.
• Anthropic’s books: $65 billion at the end of July, gross of cloud-partner sales, with an IPO path aimed at about $2 trillion.

WHAT IS UNCONFIRMED

• Year-end $70 billion: People familiar with the raise describe it as the internal target; OpenAI has not said so on the record.
• The $30 billion round: Early talks, no term sheet, value about $1.4 trillion before new money.
• Anthropic’s listing date: Expected as soon as November; the public S-1 is not yet out.

The $20 billion never left the building as lost bills. It left as a bridge between two ledgers, and the public market, which cannot buy OpenAI shares yet, marked Nvidia, Oracle and Micron instead. The raise now asks backers to treat $70 billion as a destination and $50 billion as a waypoint, on a definition Anthropic will take public first.

Disclaimer: This article is news reporting and analysis of company figures, private fundraising talks and public stock moves, and it is for information only. It is not investment advice, a solicitation to buy or sell any security, or a recommendation on OpenAI, Anthropic, SoftBank, Nvidia, Oracle, Micron or any other issuer named here. Readers should consult a licensed financial adviser or other qualified professional who can review their own objectives, risk tolerance and time horizon before acting. Run rates, valuations, deal terms and share prices are taken from the sources cited as of the dates in this piece and can change as new slides, filings or trades appear.