TL;DR · 30-second read
The Short Version
CoreWeave rents out the powerful computers that artificial intelligence companies depend on, and it is changing who pays to build them. Instead of borrowing at high interest, it expects customers to hand over about $9 billion in advance during the next three months.
Renting its computers for a short stretch now costs up to about twice as much per year as signing up for four or five years. A rival, IREN, had one customer pay 90 percent of the chip bill up front.
The upshot: the companies building artificial intelligence are increasingly paying for the computer centers themselves.
CoreWeave, a specialist AI cloud provider, reported third-quarter 2026 results on November 10 and said it expects approximately $9 billion in customer prepayments and deposits over the next three months, alongside financing backed by specific customer contracts and operating cash flow. Its revenue backlog stood at about $132.5 billion as of September 30, and it raised full-year 2026 revenue guidance to $13.0 billion to $13.3 billion.
TechCrunch reported that the quarter brought $3.66 billion in revenue, up 168% year over year, alongside a $709 million net loss and $846 million in interest expense; that CEO Michael Intrator told analysts CoreWeave will not tap the unsecured debt market at current yields; that more than $3 billion of the expected prepayments relate to capacity already built; that short-term capacity is selling for more than $40 million per megawatt a year versus $20 million to $25 million on four- and five-year contracts; and that rivals Nebius and IREN have relied on Microsoft prepayments.
Executive Summary
CoreWeave is reorganising how it pays for its buildout. Rather than issue unsecured bonds at yields that, in October, rose above 15% on some of its debt, it is leaning on three sources: loans secured against individual customer contracts, cash from operations, and money customers pay before their capacity goes live. The expected $9 billion in prepayments and deposits over a single quarter is roughly two and a half times the revenue CoreWeave booked in the whole third quarter.
The pricing data explains why customers are willing to do it. Intrator said new four- and five-year contracts are priced at $20 million to $25 million per megawatt, while short-term capacity is fetching more than $40 million. IREN’s November 5 results release reported nearly identical figures after its own quarter-end: a five-year contract at about $22 million per megawatt with 90% of the GPU capital cost prepaid by the customer, and a short-term contract at $40 million per megawatt.
For anyone buying AI compute, the implication is that a long-term commitment is no longer just a purchase of capacity at a discount. Increasingly it is also a decision to help fund that capacity, with the delivery and timing exposure that comes with paying before the hardware is running.
Customers Have Joined the Capital Stack
A company’s capital stack is the layered mix of money that funds it: shareholders’ equity, secured loans, bonds and so on. CoreWeave’s earnings release now places customers explicitly in that mix. Intrator described growth as “supported by multiple sources of funding, including financing backed by customer contracts, operating cash flow, and approximately $9 billion in customer prepayments and deposits expected in the next three months alone.” For scale, that three-month figure compares with $3.66 billion of third-quarter revenue and with the $1.05 billion to $1.15 billion of interest expense CoreWeave guides for the fourth quarter.
The composition matters. More than $3 billion is tied to new contracts for capacity CoreWeave has already built, so that cash arrives without any new hardware spending and improves its near-term position directly. The remainder supports capacity still to be built against the $132.5 billion backlog, alongside loans secured on specific contracts. IREN shows how far the model can stretch: its post-quarter five-year deal has the customer prepaying 90% of the associated GPU capital expenditure, and IREN counts prepayments within the $9 billion of cash and committed GPU financing that covers part of its $15 billion in remaining fiscal 2027 capital spending.
This moves part of the funding burden, and some of the risk, onto the buyer. A customer that prepays is financing hardware before it runs, so its exposure shifts from whether capacity will be available to whether the capacity it has paid for arrives on schedule. IREN’s own update illustrates the timing variable: its more than $4 billion annualised revenue target moved from year-end 2026 to January 2027 after a customer-requested design change and supplier timing, and may extend to February. The trade cuts both ways, though. An upfront payment also commits the provider to a price that could look cheap if demand keeps climbing.
One Megawatt, Two Prices
AI cloud capacity is increasingly quoted per megawatt, the unit of electrical power a data center can deliver to its servers, because power rather than floor space is what limits how many GPUs a site can run. On that basis, CoreWeave is running a two-tier market: more than $40 million per megawatt a year for short-term capacity, against $20 million to $25 million for new infrastructure on four- and five-year terms. That is a premium of roughly 1.6 to 2 times for flexibility.
IREN’s figures land in almost the same place, $40 million per IT megawatt (power delivered to computing equipment, excluding cooling overhead) on a short-term deal and about $22 million on a five-year one. Two providers reporting within a week do not make a market-wide index, and the measurement bases may differ slightly. Still, the convergence suggests the spread reflects current supply conditions rather than a single company’s pricing.
Both companies also report rising prices. CoreWeave raised prices across its offerings by about 25% in July and another 10% later in the quarter, with Intrator saying increases this quarter “more than offset” cost increases. IREN reported contracted pricing up 11% on three-year deals and 12% on five-year deals, and H100 GPUs contracted at double previous pricing. In that environment, a buyer locking in five years at the lower rate is paying for certainty with commitment, and increasingly with cash up front.
Lenders Still Pay for Predictability
Short-term deals earn more per megawatt, but they are harder to borrow against. Lenders weight long contracts more heavily because predictable cash flows are what repay a loan. CoreWeave’s answer is DDTL 5.5, a $2.6 billion delayed-draw term loan closed in August that it describes as its first facility designed to support shorter-duration customer contracts. Delayed-draw means CoreWeave borrows in installments as it needs the money, rather than taking it all at once and paying interest on idle cash.
Intrator’s statement that CoreWeave “will not use the unsecured debt market to fund our business at current yields” is forward-looking. During the third quarter itself, the company’s release lists an upsized convertible senior note offering of about $4.2 billion (debt that can convert into shares) and an inaugural at-the-market program to sell up to 35 million Class A shares over time. Customer prepayments add a funding line rather than replacing capital markets outright.
The structure works best while demand and pricing hold. Short contracts must be resold at renewal, and if prices soften, the premium that justifies them narrows. With fourth-quarter interest expense guided above the third quarter’s $846 million, and 2026 capital spending guided at $37 billion to $39 billion, the margin for error depends on that demand persisting.
How Buyers Should Read a Multi-Year GPU Commitment
For enterprises, AI labs and other tenants, the headline per-megawatt rate is only part of the deal. When a contract includes a large prepayment, the buyer is effectively advancing capital to its provider, so delivery milestones, remedies for late capacity, and the treatment of prepaid amounts if schedules slip deserve as much scrutiny as price.
Buyers that need certainty may still find long terms attractive, at roughly half the short-term rate. Those that can tolerate repricing gain flexibility, but at a premium that, on current disclosures, can approach double. Either way, the choice is now part procurement and part financing, and procurement teams may want treasury and finance colleagues involved earlier than they used to be.
Background
CoreWeave is a neocloud: a cloud provider built specifically to rent out GPU computing for training and running AI models. Its customers include AI labs, hyperscalers and enterprises; in the third quarter it expanded relationships with OpenAI, Cohere, Perplexity, Jane Street and others. It reported more than 1.7 gigawatts of active power and more than 4.6 gigawatts contracted, including over 1.3 gigawatts internationally in markets such as India, Indonesia and Spain.
Neoclouds have grown by building capacity ahead of revenue, financing GPUs and data centers largely with debt secured against customer contracts. IREN, which still mines bitcoin but now earns most of its revenue from AI cloud, reported $200 million in quarterly AI cloud revenue and a $20.3 billion backlog, with Microsoft and NVIDIA among its customers. As borrowing costs for the sector have risen, these providers have increasingly turned to customers to share the upfront cost. Source: CoreWeave turns to customer prepayments, shorter contracts to fund AI growth (TechCrunch): coverage of CoreWeave’s third-quarter 2026 earnings and funding strategy. Primary sources: CoreWeave Reports Strong Third Quarter 2026 Results (Exhibit 99.1 to 8-K, November 10, 2026); CoreWeave Form 8-K filed November 10, 2026; IREN Reports Q1 FY27 Results (Exhibit 99.1 to 8-K, November 5, 2026); IREN Form 8-K filed November 5, 2026; IREN Form 10-Q filed November 5, 2026.Sources