Deep Dive — OpenAI raised $11 billion of the $30 billion it wanted

OpenAI's valuation went up 47% and the cheque it attracted fell by more than half. Both numbers belong to the same round, reported the same week, and the second one is the story.

Bloomberg reported Friday that OpenAI has closed on roughly $11 billion, the first tranche of a round that values the company at $1.25 trillion, with the money wired to the company that day, according to Bloomberg. The round is not finished. It could reach about $20 billion, a source told the outlet, with the remaining $9 billion expected in the coming weeks. OpenAI declined to comment, and none of the backers were named.

Set that against the ask. In late September, Bloomberg reported OpenAI was seeking at least $30 billion at a $1.4 trillion valuation; the target was later cut to as much as $20 billion at $1.25 trillion, as Techstrong.ai summarised the reporting. Six weeks, a 63% cut to the cash, and an 11% cut to the price. The company did not get less valuable in the market's eyes. It got less fundable at size.

The comparison that makes the shrinkage legible is March, when OpenAI raised $122 billion in committed capital at an $852 billion valuation, with Amazon, Nvidia, SoftBank and Microsoft among the participants. That round had names attached. This one has a number and a wire transfer. A strategic's cheque is a procurement decision — Nvidia funding the customer that buys its accelerators, SoftBank funding the campuses it leases back. An anonymous cheque is a financial decision, and financial buyers are the ones who can wait.

The mark and the cheque are different instruments

A private valuation is a mark: an agreement about what the equity would clear at if it had to. Capital raised is a decision: how much of today's balance sheet to convert into someone else's burn. Those two can move in opposite directions without anyone behaving irrationally, and this week they did.

The mark went up because the people holding the paper need it to. OpenAI's last round priced at $852 billion in March; $1.25 trillion is 47% higher, and that number now anchors every employee tender, every secondary sale, and the floor for a 2027 listing. Nobody in the syndicate is served by marking it down three months before an IPO.

The cheque got smaller because the money converts into GPUs the moment it lands. A commitment made in September, when the Nasdaq was still near its high, is a different instrument in November, after a session that erased $2.3 trillion in market value and a midterm election that put data center restrictions on ballots in dozens of jurisdictions — the sequence we traced in Nasdaq's $2.3 trillion selloff and the 2027 capex peak. Investors did not reprice OpenAI's intelligence. They repriced the cost of being early.

There is also a cheaper way to buy the same exposure this month. Anthropic is days from a public listing. Anyone who wants AI-lab equity can wait a fortnight and buy it at a price a market sets, with a prospectus attached, instead of accepting a private mark set by the seller.

What $11 billion actually buys

The audited numbers from last year are the useful denominator. OpenAI generated $13.1 billion in revenue in 2025 against $19.2 billion in research and development spending and an operating loss of about $20.9 billion, per the financial statements Techstrong.ai cites. The tranche that closed Friday is roughly half of one year's operating loss.

Against the obligations, it is smaller still. The Information reported in July that OpenAI carries more than $665 billion in chip and data center commitments. Eleven billion dollars is 1.7% of that. It is not a war chest; it is a payment.

The burn curve makes the point sharper. The July deck the Financial Times obtained put OpenAI's negative free cash flow at $278 billion across 2026 through 2030 — the document we worked through in OpenAI's cash gap shrank while its compute bill grew. Spread evenly, that is about $55 billion a year, and the curve ramps rather than flattening, so the early years are the lighter ones. On that arithmetic, $11 billion is a quarter or two of runway, not a buildout.

Which tells you what the round is for. It is a bridge to the IPO, not the financing of the buildout. The buildout is being carried by vendors, landlords and partners — Nvidia guaranteeing a slice of the Ohio campus, SoftBank leasing the shells, Oracle and CoreWeave holding the contracts — the structure we described when Nvidia halved OpenAI's $250 billion data center backstop to $120 billion. Private equity is no longer the load-bearing layer. It is the layer that keeps the lights on until the public market takes over.

Anthropic is about to set the reference price

The Financial Times reported Friday that Anthropic is pressing ahead with a listing before Thanksgiving, targeting a raise of $80 billion to $100 billion at a valuation between $1.25 trillion and $1.75 trillion, per the Financial Times. Prospective buyers are not there. Some put the clearing value as low as $1.1 trillion, Bloomberg reported, against the roughly $2 trillion investors were said to expect a month ago, as TechCrunch noted. The Wall Street Journal reported one shareholder suggesting a figure nearer $1 trillion.

Read those two auctions side by side and the OpenAI round stops looking like a failure and starts looking like a hedge. OpenAI's private mark of $1.25 trillion sits above where Anthropic's public buyers are currently bidding. Every dollar committed to OpenAI before that auction clears is priced without a comparable. So investors commit a little, keep $9 billion unexercised, and wait for the reference price. That is not a rejection. It is a subscription with a waiting period.

The scale mismatch is the part worth sitting with. Anthropic is asking public markets for $80 billion to $100 billion in a single event — four to nine times what OpenAI just raised from private ones, from a company whose third-quarter revenue grew 1,127% year over year according to its filing. If that book clears anywhere near the low end, the "AI can't raise money" story dies, and OpenAI's problem becomes its price rather than its access. If it clears at $1 trillion or below, then OpenAI's $1.25 trillion mark is above the market, and the 2027 listing everyone is bridging toward is a down round wearing an IPO's clothes.

The permission problem became a financing problem

Thursday's selloff and Friday's shrunken cheque are the same repricing in two asset classes. Public equity moved first because it marks hourly. Private capital moved second, and it moved by size rather than price, because a private mark is sticky by construction.

The political channel is now explicit. Voters restricted data center development in nearly every jurisdiction where it appeared on a ballot, and Democrats took both chambers. OpenAI's response this week was operational rather than financial: a Grid Protectors program meant to cut the power its future campuses draw during grid stress, plus commitments to pay grid infrastructure costs and disclose water and power use, per Techstrong.ai. Those are the terms of a business that has learned its constraint is a county commission, not a fab.

The depreciation debate belongs in the same paragraph, because it is the collateral question underneath every one of these rounds. Microsoft is winding down rented capacity from Oracle after declining to renew a 2023 deal, because the hardware was built on older Nvidia chips. If a GPU's economic life is shorter than the lease written against it, then the assets backing AI's private marks are worth less than the marks assume — and the buyers who noticed that are the same buyers who wired $11 billion instead of $30 billion.

What the bears and the bulls both get wrong

The loudest bear read on Friday, from Ed Zitron, was that this is the last private money OpenAI gets before an IPO, a down round, or worse. The loudest bull read is that a target revised during a rout is not a solvency event: the company still raised at a mark 47% above March, and its revenue is growing 77% year over year, with the Financial Times putting net annualized revenue at $50 billion last quarter.

Both arguments are about the total. The structure is what changed. A round became a subscription: money wired up front, the balance optional, the backers anonymous, the price held. That is a lender's structure with an equity label on it, and it tells you the syndicate wants exposure to the outcome without exposure to the interim.

The contrarian case is that the shrinkage is, narrowly, good for the buildout. Capital raised at a lower price with a shorter leash has to go into sites that clear — and the years when money was free are precisely the years that produced 1.4-gigawatt projects a town of 20,000 could veto. A company that has to answer to public-market disclosure in 2027 is a company whose numbers can be checked. Every figure in this piece came from a leak, a filing, or a person familiar with the matter. That is the actual problem with AI's private marks, and the IPO is the only thing on the calendar that fixes it.

What to watch

The $9 billion. If the balance closes in the "coming weeks" as described, the $1.25 trillion mark holds and the round is a success with a slow draw. If it does not close, the mark is a fiction and the next event is a repricing.

Anthropic's book, before Thanksgiving on November 26. The clearing price is the number that will re-rate or confirm every private AI mark in the industry, OpenAI's included.

OpenAI's own filing. The company has said it plans to list in 2027, after delaying an earlier ambition to go public this year. Whether it files at, above, or below $1.25 trillion is the only verdict that counts.

And the fourth quarter. Fifty billion dollars annualized at 77% growth is a good business. It is not yet a business that funds $665 billion of commitments, and the gap between those two facts is what the next twelve months are about.

Is a shrunken round a warning, or just a market learning to price AI properly? Write to us via our contact page.