Michael Intrator raised the subject himself. On CoreWeave’s third-quarter earnings call on Tuesday, the chief executive told analysts that his company would stop funding itself the way it had been funding itself. “We will not use the unsecured debt market to fund our business at current yields,” he said, “nor do we need to.”
The quarter explains the sentence. CoreWeave reported $3.66 billion in revenue, up 168% from a year earlier and above the top of its own forecast. It also posted a net loss of $709 million and $846 million of interest expense, against a contracted backlog of $132.5 billion. Those interest payments are what the new financing plan is designed to shrink.
The plan has three parts. CoreWeave expects about $9 billion in customer prepayments and deposits over the next three months, including more than $3 billion tied to signing new contracts for capacity it has already built and therefore does not have to fund again. It is borrowing against specific customer contracts rather than against its own signature, drawing on facilities as it spends; in August it closed a $2.6 billion loan, its first designed around shorter-term contracts. And it is selling more of its capacity on short contracts at much higher prices.
That last part is where the margin is. Intrator said short-term deals are clearing at more than $40 million per megawatt annually, while new capacity on four- and five-year contracts is priced at $20 million to $25 million per megawatt. The company raised prices about 25% in July and another 10% later in the quarter, and said prices kept climbing in the current period.
A prepayment is revenue pulled forward, and it is also a promise. The cash arrives now; the obligation to deliver computing at a price agreed now arrives over the life of the contract. If the price of artificial-intelligence compute keeps rising, CoreWeave will have sold forward at the wrong number, and the customer will have bought an option on the company’s own cost inflation.
That is the mirror image of the debt it is avoiding. Bondholders took a view on the risk last month, when yields on some of CoreWeave’s bonds rose above 15%. Equity holders took a different one this week, sending the stock up as much as a quarter on Wednesday as the market read the results and the financing plan as a reduction in refinancing danger. Two markets looked at the same contracts and reached opposite conclusions about how much they like them.
The difference is collateral. A four-year contract with a creditworthy customer is an asset a lender can underwrite, and lenders prefer it even at a lower price because the cash flow is easier to predict. Short contracts that reprice upward are better for the seller and worse for the bank. The new facilities resolve the mismatch by taking the contract itself as security, so the loan matures with the customer rather than with the company’s intentions.
There is precedent on the demand side. Nebius and IREN have both funded capacity with large prepayments from Microsoft, part of a broader shift in which the biggest technology companies are becoming the balance sheet of the industry that serves them. A buyer that pays three years in advance is, in effect, lending its credit rating to a smaller supplier in exchange for guaranteed capacity at a locked price.
CoreWeave’s version is more crowded. It is asking hundreds of AI developers, not one software company, to pay in advance, and its backlog will now be judged on how much of the $132.5 billion is prepaid, how much is contracted with customers whose own funding is secure, and how much carries cancellation terms. The company said demand for its 2027 capacity exceeds what it can deliver by ten to one, which is the strongest argument for asking for cash up front. It is also an argument that is only as good as the customers making it.
Then there is the renewal question. A short contract lets a seller reprice in twelve months if demand holds, and forces it to find a new customer in twelve months if it doesn’t. CoreWeave is trading duration for price near the top of a cycle, which is a reasonable trade for a business that believes compute is still underpriced and a dangerous one for a business that is wrong about the cycle.
The raised guidance helps the case. CoreWeave lifted its full-year revenue forecast to a range of $13.0 billion to $13.3 billion and described demand as contracted rather than projected. But guidance is an estimate and a prepayment is a liability, and the two sit on different lines of the same page.
What the quarter establishes is narrower than the stock move. CoreWeave has found a way to grow without issuing the debt that made its bonds trade like a startup’s, and it has persuaded its customers to fund the difference. Whether that is a permanent change in how AI infrastructure gets financed, or a structure that works only while customers fear missing capacity more than they fear paying early, is what the next two quarters will answer.
