CoreWeave flagged for heavy cash burn

CoreWeave’s problem is not that demand looks weak. It is that demand looks enormous, and serving it may be ruinously expensive. The AI cloud company said on February 26 that 2025 revenue reached $5.13 billion, up from $1.92 billion a year earlier, and that its revenue backlog had climbed to $66.8 billion. It also told investors to expect $12 billion to $13 billion of revenue in 2026. Those are giant numbers. They are also the numbers that make the rest of the story easy to miss. CoreWeave lost $1.17 billion in 2025 under standard accounting, paid $1.23 billion in net interest expense, and now plans to spend $30 billion to $35 billion on capital expenditures in 2026 after spending $10.31 billion in 2025 (investors.coreweave.com; cnbc.com). That is where the “$2.60 for every $1 of revenue” critique comes from. If CoreWeave hits the midpoint of its 2026 revenue guide, about $12.5 billion, and the midpoint of its capex plan, about $32.5 billion, it will be pouring in cash at a rate that overwhelms its sales. A backlog does not change that arithmetic. Backlog is not cash in the bank. CoreWeave itself defines the figure broadly. It includes remaining performance obligations plus other amounts it estimates will be recognized later under committed customer contracts, and even then only if delivery and service availability requirements are met (investors.coreweave.com; sec.gov). The company’s whole model explains why the spending is so intense. CoreWeave is not a software business with light infrastructure and fat margins. It is a financing machine wrapped around Nvidia chips, power contracts, and data center buildouts. In its 2026 outlook it said it wants to finish the year with more than 1.7 gigawatts of active power capacity, up from 850 megawatts at the end of 2025. That means more sites, more hardware, more debt, and more time spent racing construction schedules. When CoreWeave missed investors’ hopes in February, the immediate trigger was not weak demand. It was the scale of the buildout and the losses attached to it (cnbc.com; bloomberg.com). That financing burden was visible long before the latest warning. In its IPO filing, CoreWeave said the vast majority of its revenue came from multi-year committed contracts, often on a take-or-pay basis. That sounded reassuring. The same filing also showed how concentrated the business was. Microsoft was already central enough that CoreWeave later told the SEC there had been no cancellations of its committed Microsoft contracts after regulators asked. By 2026, OpenAI had become another huge pillar. CoreWeave’s 2025 annual report said OpenAI had committed to pay up to about $6.5 billion under one order form through May 2031, and the company expected OpenAI to become a significant customer (sec.gov; sec.gov; sec.gov). That concentration matters because AI infrastructure looks safer than it is when everyone focuses on backlog and ignores who is actually supposed to pay it. A contract with a hyperscaler or a model lab is better than a handshake with a startup. It is still not the same thing as durable profitability. CoreWeave’s 2025 results show the gap clearly. Revenue exploded. Adjusted EBITDA looked strong. Yet operating expenses still exceeded revenue for the full year, and interest expense alone ate nearly a quarter of annual sales. In the fourth quarter, interest expense was $388 million on $1.57 billion of revenue, while net loss widened to $452 million (investors.coreweave.com; cnbc.com). This is the part boards and investors keep relearning in AI. Demand can be real and the business can still be fragile. CoreWeave may yet grow into its footprint. Nothing in the current numbers proves it cannot. But the latest filings do prove what kind of company it is right now: one that turned $5.13 billion of 2025 revenue into a $1.17 billion net loss, then answered with a 2026 capex plan as high as $35 billion and a promise to more than double active power capacity by year end (investors.coreweave.com; cnbc.com).