What happened
CoreWeave’s November 10, 2026 announcement reports results for the quarter ended September 30. Revenue was $3.661 billion, compared with $1.365 billion in the same quarter of 2025. Operating income was $207 million, up from $52 million. Net loss was $709 million, versus $110 million a year earlier; net interest expense was $846 million. The company reported adjusted EBITDA of $2.256 billion, with a 62% adjusted EBITDA margin, and adjusted operating income of $379 million, with a 10% margin. Those adjusted measures are non-GAAP and are supplemental to its GAAP results.
Revenue backlog stood at approximately $132.5 billion on September 30. CoreWeave defines this as remaining performance obligations plus other amounts it estimates will be recognized under committed customer contracts. The company says those amounts remain subject to delivery and service availability requirements. That qualification matters: backlog is a forward-looking measure tied to execution, not cash already collected or revenue already recognized.
The company said it added more than 200 megawatts of active power during the quarter, bringing its active footprint above 1.7 gigawatts and doubling that footprint since the start of the year. Total contracted power rose by more than 900 megawatts to over 4.6 gigawatts. CoreWeave said its contracted power outside the United States exceeded 1.3 gigawatts, including in India, Indonesia and Spain. Its year-end expectation is more than 1.85 gigawatts of active power.
CoreWeave raised its 2026 revenue guidance to $13.0 billion–$13.3 billion and adjusted operating income guidance to $1.275 billion–$1.375 billion. It expects 2026 capital expenditures of $37 billion–$39 billion, saying it expects to deploy infrastructure and deliver it to customers sooner than planned. The company also raised its expected exit-year annualized run-rate revenue to $20 billion–$21 billion; it calculates that metric by multiplying the final month of a quarter’s revenue by 12. That is a run rate, not a forecast of recognized full-year revenue.
The announcement also describes product and infrastructure developments. CoreWeave launched Forge, an AI development environment integrating training, inference, evaluation and agent development workflows. It said production workloads for Cognition on multi-rack NVIDIA Vera Rubin NVL72 deployments achieved 4.8 times the token throughput of GB200 NVL72. CoreWeave also announced support for NVIDIA Vera CPUs. These are company-reported milestones; the announcement does not establish that all customers will see the same performance.
What the archive adds
New Runtime’s earlier coverage on October 13 described how OpenAI revenue details affected market readings of AI infrastructure exposure, before any change in product demand was established (Nvidia, Oracle and CoreWeave shares fell following OpenAI revenue details). This is a new company-results event that adds operating figures and raised guidance to that demand-and-valuation context. New Runtime also covered a hardware supplier’s record sales and capacity constraints on October 19 (ASML posted record Q3 sales and raised guidance); that is adjacent context for the broader buildout, not evidence about CoreWeave’s own delivery.
Trend signal
This does not yet fit a pattern New Runtime tracks. CoreWeave’s results speak to capital intensity, committed demand and power availability in AI infrastructure, but none of the tracked patterns specifically captures cloud-provider financial performance or capacity expansion. The reported backlog and contracted power are useful indicators to watch, but one company’s results do not establish a general market trend or demonstrate that capacity can be delivered on schedule.
Why now
CoreWeave says it raised its capital-expenditure outlook because it expects to deploy infrastructure and deliver it to customers sooner than planned. Its results therefore arrive alongside a concrete acceleration in its own buildout, including more than 200 megawatts of added active power in the quarter. The announcement connects that timing to expected customer deliveries, rather than describing a general change in the market’s timetable.
Connections to other events
The results connect to New Runtime’s earlier coverage of OpenAI revenue details and their effect on how investors read AI infrastructure exposure. That earlier note cautioned that market interpretation could shift without evidence of changing product demand; CoreWeave now reports its own demand-linked backlog and raised guidance, but its backlog remains contingent on delivery and service availability. Its expanded power footprint also sits alongside New Runtime’s earlier coverage of ASML’s capacity outlook, though equipment supply and cloud-provider capacity are distinct constraints.
Why it matters
Teams buying compute should read the numbers as context about the scale and commitments behind AI cloud supply, not as a promise that capacity will be available to any particular customer. CoreWeave’s backlog is large, but the company’s definition explicitly depends on future delivery and service availability. Its more than 4.6 gigawatts of contracted power and above 1.7 gigawatts of active power also show that contracted resources and operating capacity are different measures.
The financial picture is mixed. Revenue and operating income grew, but CoreWeave recorded a GAAP net loss and $846 million in quarterly net interest expense while planning $37 billion–$39 billion in annual capital expenditure. Those figures matter to customers assessing supplier resilience, investment horizons and the risk of depending on a single provider. They do not establish that CoreWeave faces imminent financial distress, nor do they show how its costs compare with competitors.
Forge and the announced hardware milestones may matter to builders evaluating integrated environments for training, inference, evaluation and agent workflows. But the announcement does not provide a general price comparison, access terms or independent performance results that would settle a procurement decision. Teams should distinguish the company’s stated throughput result on a specific deployment from performance they can expect on their own workloads.
What happens next
- CoreWeave will report whether active power exceeds 1.85 gigawatts by year-end. (3 months, 75%). Watch for: its next results or filing disclosing active power at December 31.
- CoreWeave will report 2026 revenue within its raised $13.0 billion–$13.3 billion range. (3 months, 65%). Watch for: audited annual results and any reconciliation of guidance to reported revenue.
Limits and uncertainty
These are company-reported results and guidance, not an independent assessment of customer demand, infrastructure quality or future financial performance. Backlog includes estimated amounts and is conditional on delivery and service availability; it should not be treated as guaranteed revenue. CoreWeave’s guidance and year-end power expectation are forward-looking statements and may not materialize. Its adjusted EBITDA, adjusted operating income and annualized run-rate revenue use company-defined non-GAAP or operational measures; comparisons with other providers may not be like-for-like. The 4.8-times throughput claim is tied to specified production workloads and hardware, and the announcement does not establish that it generalizes across customers or tasks.
Skill to build now
When assessing an AI cloud provider, build a small capacity-and-commitment checklist this week: record active capacity separately from contracted capacity, backlog separately from recognized revenue, and GAAP results separately from adjusted measures. For each supplier, note the delivery conditions attached to commitments and ask which workload, hardware and service level a performance claim describes. This helps teams avoid treating a large backlog or a benchmark-like result as a guarantee for their own deployment.
Reader takeaway
CoreWeave’s quarter shows fast growth and an enormous reported backlog, but also heavy investment, significant interest expense and ongoing delivery conditions. For AI builders, the useful signal is not simply that demand is strong: it is that turning commitments into usable compute still depends on capital, power and execution.