When CoreWeave reported its Q1 2026 results on May 7, the headline numbers looked like a company under pressure. Revenue of $2.078 billion, more than double the $982 million it posted in Q1 2025. Net loss of $740 million, up from $315 million a year earlier. Interest expense of $536 million for the quarter, against $264 million in the same period last year. Shares fell as much as 10% in after-hours trading, as investors reacted to light second-quarter revenue guidance rather than the Q1 results themselves.

Reading those figures in isolation produces the wrong conclusion.

The number that makes the rest of the income statement legible sits in the cash flow statement: CoreWeave spent $7.695 billion on property and equipment in Q1 2026, compared to $1.407 billion in Q1 2025. That figure is the cost of building data centers, GPU clusters, and power infrastructure before customers pay for them. The $536 million in quarterly interest expense is the carrying cost of the debt raised to fund that buildout. CoreWeave’s losses are the accounting consequence of a deliberate decision to build physical infrastructure at maximum speed.

The demand side of that bet is quantified in the revenue backlog: $99.4 billion as of March 31, 2026. That figure represents committed customer contracts whose revenue has not yet been recognized. At $99.4 billion, it suggests the infrastructure being built has buyers waiting for it.

The Customer Roster

The Q1 customer wins make the backlog less abstract. CoreWeave executed multiple new agreements with Meta during the quarter, including a $21 billion commitment signed in March. It signed a multi-year agreement with Anthropic to support the development and deployment of the Claude family of models. It expanded relationships with Cohere, Jane Street, Mistral, Perplexity, and others. Simultaneously, NVIDIA closed a $2 billion equity investment in CoreWeave, and the company secured an $8.5 billion delayed draw term loan facility to fund continued expansion. CoreWeave also surpassed one gigawatt of active power during the quarter, with a stated target of more than eight gigawatts by 2030.

What the Non-GAAP Figures Show

The non-GAAP figures add important texture to the GAAP losses. Adjusted EBITDA for Q1 2026 was $1.157 billion, representing a 56% margin on revenue of $2.078 billion. That margin contracted from 62% in Q1 2025, which reflects the scaling costs absorbed during the period. Adjusted EBITDA strips out depreciation, amortization, interest, taxes, and stock-based compensation, the items most directly driven by the infrastructure buildout, and what remains is a business generating substantial cash from its existing capacity. The question the GAAP losses pose is not whether the underlying business works. It is whether the pace of expansion can be sustained without the financial structure becoming unmanageable.

The Balance Sheet

On that question, Q1 2026 provides evidence in both directions. Current debt stood at $7.547 billion and non-current debt at $17.312 billion as of March 31, against total assets of $55.573 billion, of which $36.424 billion was property and equipment. The company generated $2.984 billion in operating cash flow during the quarter, a significant increase from $61 million in Q1 2025, which reflects the scale at which existing infrastructure is now being monetized. Cash and cash equivalents stood at $2.244 billion at quarter end, down from $3.127 billion at the start of the year, with the difference largely explained by the gap between capital expenditure and financing proceeds received during the period.

The Thesis

CEO Michael Intrator, in the company’s earnings release, framed the quarter around a specific competitive thesis: that as the AI market shifts from training to inference, the value of purpose-built infrastructure increases. What the filing verifies is that the companies making the largest AI bets, Meta, Anthropic, and NVIDIA, are backing that thesis with capital and long-term commitments.

Whether CoreWeave is winning the AI infrastructure race is, at this stage, the wrong question. The race has not reached a point where a winner can be declared. What Q1 2026 establishes is that CoreWeave has structured itself to compete at a scale and speed that requires believing AI infrastructure demand will be both large and durable. The $99.4 billion backlog, the $21 billion Meta commitment, and the $7.695 billion in single-quarter capital expenditure are not the numbers of a company hedging that belief.