CoreWeave’s Nov. 10 results delivered the kind of numbers that make the AI-infrastructure boom look both validated and financially precarious at the same time. Third-quarter revenue reached $3.661 billion, up from $1.365 billion a year earlier. The company raised full-year revenue guidance to $13.0 billion to $13.3 billion and said reported backlog stood at about $132.5 billion at Sept. 30. But the same quarter also carried $846 million of net interest expense, a $709 million GAAP net loss and a newly lifted 2026 capital-expenditure range of $37 billion to $39 billion.
That contrast is why these results matter beyond one stock move. CoreWeave is becoming a test case for whether AI cloud can be financed largely through customer demand, prepayments and asset-backed borrowing before the bills for power, data-center buildout, GPUs and interest overwhelm the economics. The real question is not whether customers want AI compute. It is whether that demand can be turned into durable cash returns.
Demand is not the missing piece
On the surface, the quarter answered the easiest part of the debate. Demand looks strong. CoreWeave reported operating income of $207 million and adjusted operating income of $379 million, while lifting its exit-2026 annualized run-rate revenue forecast to $20 billion to $21 billion. It said active power topped 1.7 gigawatts after adding more than 200 megawatts during the quarter, and contracted power rose above 4.6 gigawatts. More than 1.3 gigawatts of that contracted power sits across international markets including India, Indonesia and Spain.
The company is also trying to show it is more than a GPU landlord. It launched CoreWeave Forge as an AI development environment spanning training, inference, evaluation and agent-development workflows. It announced multi-rack NVIDIA Vera Rubin NVL72 deployments and said first production workloads for Cognition on that platform achieved 4.8 times greater token throughput than GB200 NVL72. That performance comparison is CoreWeave’s own reported result for a specific workload, not a universal benchmark, but it points to the company’s strategy: sell a fuller stack and move up the value chain.
Investors clearly paid attention. The Associated Press reported CoreWeave shares jumped 24.7% after the release.
Still, the backlog figure needs to be read for what it is. CoreWeave defines it as remaining performance obligations plus other amounts it expects to recognize under committed contracts, subject to delivery and service-availability conditions. In other words, it is evidence of demand, not a cash balance and not guaranteed profit. A cloud customer may be committed on paper, but CoreWeave still has to bring power online, install and operate equipment, and meet service terms before the revenue fully counts.
The cash waterfall is the story
The most important number in the quarter may not be backlog or revenue, but the gap between operating momentum and the financing burden underneath it.
CoreWeave posted adjusted EBITDA of $2.256 billion, a 62% adjusted-EBITDA margin. That sounds enormous, and it helps explain why lenders and equity investors have been willing to fund expansion. But adjusted EBITDA sits above some of the costs that define whether this business model can endure. Net interest expense alone was $846 million, up from $310 million a year earlier. GAAP net loss widened to $709 million from $110 million.
That gap is the AI-cloud financing loop in plain view. The company raises capital, buys or leases specialized hardware, secures data-center capacity and power, then rents that compute to customers. If utilization is high and pricing holds, revenue can scale quickly. But the money has to clear a long waterfall: power and facilities, networking and storage, maintenance and support, depreciation and amortization, interest, principal repayment and eventually refresh capex when a newer accelerator generation changes the performance-per-dollar equation.
This is why CoreWeave’s 2026 capex outlook matters so much. A $37 billion to $39 billion buildout is not just growth spending; it is a race to keep enough capacity arriving on time for contracted demand. Delays in power, construction or hardware delivery can slow revenue recognition even while financing costs continue to accrue. In a business where the assets are expensive and technologically perishable, timing is part of the margin structure.
How CoreWeave is trying to close the gap
Management’s answer is to make customer commitments do more of the funding work.
CoreWeave said it expects about $9 billion of customer prepayments and deposits in the next three months. On the earnings call, management said more than $3 billion of that amount is tied to existing cloud infrastructure that requires no additional capital expenditure. The company has also added financing channels meant to match asset lives and contract terms more closely: an upsized convertible senior notes offering that raised about $4.2 billion, an at-the-market equity program for up to 35 million Class A shares, and a $2.6 billion DDTL 5.5 facility intended to support shorter-duration customer contracts. Management said it expects to launch another delayed-draw term loan facility and expects that facility to finance more than 90% of associated capex. Under current plans, it said, it does not intend to issue additional parent-level unsecured debt through 2027.
Strategically, that matters. If more of the buildout can be financed at the asset level and anchored by customer commitments, CoreWeave becomes less dependent on repeated corporate-level borrowing. The company’s CoreWeave Omni offering points the same way. By letting customers run workloads on third-party infrastructure through the CoreWeave platform, Omni offers a path to capture demand with less balance-sheet intensity than owning every server deployment outright.
But this is also where the unanswered questions live. The public record does not show how much of backlog sits with investment-grade customers, how much capacity is already supported by operating infrastructure, what the cancellation or minimum-use terms look like, or how refundable those expected deposits may be. It also does not show the full hardware return model: utilization by GPU generation, realized revenue per unit of compute, depreciation schedule, residual values or refresh timing.
For customers buying capacity, the practical lesson is that cloud reliability is now partly a credit question. A buyer should care not only about access to GPUs, but also about delivery milestones, liquidity, service-level remedies, refresh plans and workload portability if a build slips. For investors and lenders, the discipline is different but related: separate reported backlog from collected cash, adjusted profitability from GAAP earnings, and contracted demand from capacity already online.
CoreWeave’s quarter made the bullish case for AI infrastructure easier to see. It did not settle the harder question of whether that demand can finance itself all the way through the capital stack. For now, the company appears to be winning the market-share race and still underwriting an enormous construction and refinancing challenge at the same time.
By
By
By
By
By
By
