Deep Dive — OpenAI's $30B round fell to $20B at a $1.25T valuation
In September, OpenAI went to investors asking for $30 billion at a $1.4 trillion valuation. On Thursday night the Financial Times reported the target is now about $20 billion at $1.25 trillion, because some investors cut the size of their commitments amid the AI selloff, according to the FT. Hours earlier, the Wall Street Journal had put a number on how far the round has actually got: roughly $11 billion committed, less than half the original target, the Journal reported.
That is the single most important number in AI this week, and it is not a demand number. It is the price of money.
Two reports from two papers on the same evening, agreeing on the valuation and disagreeing on the size, is what a market looks like when it is being repriced in public. The round was supposed to be a bridge: a private raise to carry OpenAI to an IPO it had already pushed past 2026. A bridge that comes in at half its target is not a crisis, but it does mean the company has to choose between raising again sooner, raising at a lower number, or spending less than it told its compute suppliers it would.
What OpenAI is actually raising
The starting point was clean. In late September, Bloomberg reported OpenAI was in talks to raise at least $30 billion at a valuation of roughly $1.4 trillion, explicitly as a pre-IPO bridge round. The company's run-rate revenue had jumped about 70% since July to $40 billion in August. It last priced at $852 billion, and an August tender offer for employee shares came in at the same mark, which tells you the private market had not moved on it in months.
Then the arithmetic changed from both ends. The valuation came down about 11%, from $1.4 trillion to $1.25 trillion. The raise came down a third, from $30 billion to $20 billion. And the money actually in the door — $11 billion — is a little over half of the revised target and roughly a third of the original one.
The FT's account adds a detail that matters more than either headline: the first close is expected to land below $20 billion, with a later offering used to fill the gap. That is the vocabulary of a round being stitched together in tranches rather than subscribed in one go, and it is the same structure Bloomberg reported on Thursday for Baseten, which we will come back to.
It is also worth separating what is reported from what is known. Both stories rest on people familiar with the matter, not on anything OpenAI has said. A target is not a result, and a round that closes at $11 billion in October can close at $20 billion in December. The Journal itself says OpenAI expects to raise more in the coming weeks. What has changed is not the amount of money in the world. It is the number of investors willing to write a cheque at the old price.
Meta is the same story with a credit rating attached
On Friday, S&P Global Ratings moved Meta's outlook from stable to negative, keeping the AA- rating but flagging "uncertainty" around elevated capital expenditure and free cash flow, CNBC reported.
The specifics are worth reading slowly, because they explain why a ratings agency is involved in an AI story at all. Meta guided full-year capex to $135 billion–$145 billion. Last quarter it posted negative free cash flow of $2.12 billion — the first time in the company's history it has gone into the red on that measure. Shares fell 7% on the print despite a revenue beat, and finished October down 15%, the worst month in four years.
On Monday, Meta said it plans to raise $95 billion, starting with $35 billion of underwritten public equity and following with $60 billion through an at-the-market programme. Read the two announcements together and the sequence is clear: the cash flow went negative, the equity raise was announced, and four days later the outlook went negative. S&P is not objecting to AI spending in the abstract. It is objecting to a company whose capex now exceeds what its operations generate, which is a description of every large AI spender at some point in the next eighteen months.
Meta can absorb this. It has the ad business, the rating is still investment grade, and an outlook change is not a downgrade. But an outlook is a warning shot with a date on it, and the trigger is mechanical: another quarter of negative free cash flow, or a capex guide that goes up again, and the rating follows the outlook.
The bond window closed before the equity window did
The neoclouds got there first, and we know how that ended.
In recent weeks, CoreWeave, Nebius and Nscale have effectively been shut out of the bond market as lenders and investors demand tougher terms, per the Wall Street Journal. That is a sentence about collateral, not sentiment. The same month, Bloomberg reported that a CoreWeave-tied data center raised $1.1 billion of junk bonds at 98.5 cents on the dollar to yield 9.25% — about 2.7 percentage points above similarly rated debt. A spread like that is the market saying it does not believe the rental contract behind the building will survive a downturn intact.
The equity market has already voted on the same companies. CoreWeave, Nebius and IREN had each fallen more than 30% in October as of last week, CNBC reported. We traced the mechanism five days ago in how the neoclouds hit a credit wall while SpaceX borrowed $40 billion: the lenders did not stop liking GPUs, they stopped accepting a GPU rental contract as collateral for a twenty-year building. Japan's financial regulator reached a similar conclusion in September, when it put AI data center loans under scrutiny — one of the first supervisors to treat the sector's debt as a category rather than a client.
So the chip vendor became the lender
Which brings us to the most structurally interesting story of the week, and the one that will still matter in 2028.
Nvidia is in talks to guarantee billions of dollars of debt financing for Nscale's planned Anthropic data center in West Virginia, The Information reported, with Bloomberg corroborating. The instrument is a residual value guarantee: if Nscale defaults on construction payments, Nvidia steps in, takes over the facility temporarily, and helps find a buyer and a new tenant. Nvidia's previous support for cloud providers covered leases and chip purchases. This one covers construction.
The reason Nvidia would do it is visible in the contract underneath. Anthropic has signed a $44.6 billion deal for the first building alone — 460 megawatts, built for Nvidia's next-generation Vera Rubin chips — according to Nscale's IPO filing. A second building, worth roughly another $45 billion, is under discussion.
Nscale is trying to go public at up to a $35 billion valuation, has about 25,000 advanced AI chips running against a plan for nearly 500,000, and watched Microsoft pull out of a planned English site this month when grid power slipped into the 2030s. The project needs a creditworthy backstop. The chipmaker needs the megawatts to exist, because they are the only place its next generation gets deployed at that scale.
Nvidia has already extended roughly $170 billion in investments and credit support to neoclouds and large customers, per The Information, including $1 billion of convertible financing to Nscale in September. At some point the distinction between selling someone a data center's worth of accelerators and underwriting the building that holds them stops being a commercial relationship and becomes a balance-sheet position. Nvidia is now, functionally, the lender of last resort for the AI build-out in the United States.
The rating agencies are watching the same seam. Anthropic filed to go public this week, and Blackstone has warned investors that it may receive a below-investment-grade rating after listing, per TechCrunch. A junk rating raises Anthropic's borrowing cost and pushes it back toward exactly this arrangement: a stronger partner standing behind its leases, so the paper can be sold at all.
What still gets funded, and why that is not a contradiction
On the same day OpenAI's round was reported shrinking, Bloomberg reported that Baseten is raising about $2 billion in two tranches, one valuing it at $26 billion and the other at $30 billion. Baseten was valued at $13 billion in June, after raising $1.5 billion. That is a doubling in four months.
The two stories are not in conflict, and the reason is the shape of the money rather than the mood of the market. Equity in an inference provider is a claim on gross margin and usage: if demand grows, the equity owner participates, and if the build slips a quarter the equity simply waits. Debt on a building is a claim on a tenant paying rent on a schedule. Equity can absorb delay. Debt cannot, which is why the debt market moved first and hardest, and why the equity market is still writing cheques — at a lower price for the largest private company in the sector, and a higher one for the fastest-growing smaller ones.
There is a second reason Baseten can raise at 2x while OpenAI's round shrinks. Inference is where the revenue actually lands, and it is where the margin is thinnest for everyone else. The same logic explains why Foxconn expects to hold half the AI server market at a 6.12% gross margin, which we looked at in Foxconn's 50% share at a 6% margin: the money in this build-out collects at whichever point in the stack is scarce, and right now that is serving tokens, not pouring concrete.
The skeptic's read
Three readings of the week are available, and it is worth being honest about which one the evidence supports.
The first is that demand is breaking. The evidence for it is weak. OpenAI's run-rate revenue was $40 billion in August and approaching $50 billion by the end of September, per the FT, and Meta beat on revenue. The companies cutting commitments are investors repricing risk, not customers cancelling subscriptions. Anyone who told you in 2025 that AI demand would collapse is still waiting.
The second is that the collateral is being repriced, which is the reading this column takes. The market is no longer asking whether AI compute will be rented; it is asking who holds the residual value if it is not. For a neocloud, that answer is now "nobody, at any price we like." For Meta, it is "the equity holders, $95 billion of them." For Nscale, it is "Nvidia." Each of those is a different instrument solving the same problem, and each of them transfers risk toward the party with the strongest balance sheet — which is a description of a market maturing, not of one collapsing.
The third reading is that nothing was ever disciplined and the froth has simply changed instruments. Baseten doubling in four months is the strongest evidence for it. A company raising at $30 billion eleven days after its last mark at $13 billion is not a correction; it is the same enthusiasm pointed at a different ticker. If that is right, the repricing we are watching is a rotation, and the next leg down starts wherever the newest money landed.
The honest uncertainty is this: both of Thursday's OpenAI reports are sourced, not confirmed. If the round closes at $11 billion against a $20 billion target, the shortfall does not vanish — it becomes a smaller compute bill, later, which lands on the backlogs of the companies that signed contracts expecting OpenAI's bridge to fund them. That is the transmission channel worth watching, and it runs from a term sheet in San Francisco to a bond spread on a data center in Texas.
What to watch
The first close. The size and the date of OpenAI's initial close will leak within weeks. Under $15 billion would mean the revised target is itself already stale.
Meta's next quarter. S&P has set up a mechanical test: free cash flow, and whether the capex guide moves again. A downgrade from AA- would raise the cost of the $95 billion it is trying to raise.
Any neocloud bond deal that prices. The clearest sign the window has reopened is a CoreWeave or Nebius issue that clears at a spread under 5 points. Nothing else in this story means as much.
Whether the Nvidia guarantee becomes a template. If residual value guarantees spread to other cloud customers, the chipmaker is underwriting the industry's construction risk, and its earnings stop being a clean read on chip demand.
Anthropic's rating at IPO. A junk rating makes partner-backed debt the default financing structure for frontier labs, which is precisely the arrangement Nvidia is negotiating now.
Is OpenAI's smaller round a warning about AI demand, or just the price of money going up? Write to us via our contact page.
