Sarah Friar was asked on CNBC whether the midterm elections would change how OpenAI builds data centers. The chief financial officer said they would not. Political results move on one calendar; the company’s commitments to land, power and chips are written on another, and her answer made that separation explicit.
By the same afternoon a larger number had arrived. Bloomberg reported that OpenAI has raised $11 billion in a new round at a valuation of $1.25 trillion. People familiar with the matter said the financing has not closed and is expected to wrap within weeks, and that the total could reach about $20 billion.
The comparison that matters is to September. Bloomberg reported then that the company was seeking $30 billion. A round that has banked $11 billion, may finish near $20 billion and was once described at $30 billion is what a very large private financing looks like while it is still being assembled. Commitments arrive in tranches, and the headline figure at any moment counts only the ones that have been signed.
For OpenAI the money has one obvious destination. The company has spent two years converting capital into computing capacity, contracting for data centers, buying accelerators and securing electricity, and the pace of that spending is what its finance chief was defending on television. An $11 billion installment buys time on that schedule. It does not alter the schedule.
At $1.25 trillion the valuation outranks all but a handful of listed companies, and it is being set by private buyers rather than by a public book. That is a different kind of number. It reflects what a small group of investors agreed to pay for a stake they cannot sell on a screen, in a company whose spending commitments are visible only in fragments.
The listing question hangs over the round. Earlier this year OpenAI signaled it might go public, then let the timetable slip. Investors now expect an offering next year. If that happens, $1.25 trillion becomes the price the company has to defend in a filing, with audited numbers attached, in front of buyers who can compare it against every other asset they could own.
The buyers in this round have not been named publicly. In financings of this size the identity of the investor often matters as much as the amount. Strategic holders can bring cloud capacity, chip allocations or distribution along with their cash, and they can accept terms a purely financial buyer would refuse. A price agreed in that setting is negotiated, not discovered.
The rest of the market spent the week moving the other way. CoreWeave’s corporate bonds traded at yields above 16%, and CNBC described the debt as the credit card of the AI buildout, one whose statement is getting more expensive. SoftBank’s five-year credit default swaps widened to roughly 466 basis points, a four-year high, and its shares closed down 11% in Tokyo at 3,549 yen. Equity for the largest lab is plentiful. Debt for the infrastructure beneath it is not.
That split has a mechanical explanation. Compute rented from someone else ages with each chip generation, and the equipment inside a data center depreciates on a schedule measured in a few years, not decades. Lenders price that decay. Equity investors in the model developer price the revenue the machines are supposed to produce, which is a longer and less certain story. When the two prices move apart, the gap lands on the smaller firms in the chain as a higher cost of capital.
There is a simpler reading of the $11 billion figure. OpenAI may have chosen to take less money at a higher price, limiting dilution while the valuation is strong, rather than raise the full amount it once floated. Under that reading the September target was a ceiling rather than a commitment, and the slide from $30 billion to $20 billion is discipline rather than weak demand.
The other reading is less flattering. If the round closes at $20 billion, the company will have raised two-thirds of what it sought six weeks earlier, at a price its own investors set. Buyers already exposed to the same buildout, through bonds, leases or supplier contracts, have reasons to be careful about adding more at the top of a cycle.
Friar’s answer about the midterms is the part suppliers will read most closely. Data center contractors, turbine makers and chip vendors plan against multiyear orders, and those orders survive elections only if the buyer’s balance sheet does. Her statement put the company’s construction plans outside politics. It did not put them outside financing.
There is a third possibility that the round’s structure leaves open. A financing that is described as $20 billion but reported at $11 billion may include commitments that are contingent on capacity being delivered, and capacity depends on power contracts that have their own schedules. Money pledged against future compute is not the same as money in hand, which is why the September number and this week’s number can both be accurate descriptions of the same negotiation at different moments.
For the suppliers the practical question is which number they can plan against. A vendor that tooled up for a $30 billion program and now sees $11 billion committed has an inventory problem, and the same vendor will be asked to hold capacity for the next round on the strength of a valuation it cannot audit.
What happens in the next few weeks is narrower than the headline. The round either closes nearer $11 billion or nearer $20 billion, and the buyers who set the mark either stay for a public listing or take their discount and leave. Both outcomes are compatible with a $1.25 trillion price on paper, which is the only place that price currently exists.
