CoreWeave Doubles Revenue to $2.58 Billion but Loss Jumps 116% with Debt and Capex of $9.4 Billion
GPU rental provider concluded the quarter with $104 billion in backlog and a new $21 billion contract with Meta through 2032. The size of the debt has become the new debate.
CoreWeave, one of the leading providers of GPU rental capacity in the market, closed the second quarter of 2026 on Tuesday night, August 11, with revenue of $2.575 billion, up 112% year-over-year and slightly above the $2.56 billion consensus expected by sell-side analysts. However, the point that dominated analysts' focus was not the revenue beat but rather the gap between contracted volume and financial results: net loss jumped 116% to $626 million, compared to $290 million a year earlier.
The engine of the loss is debt. The company's net financial expenses for the quarter reached $640 million, more than double the $267 million recorded in the same period of 2025. The company spent $9.4 billion on capex alone in the second quarter, raising its spending forecast for the entire year, a move that further stretches the balance sheet at a time when the cost of capital remains high.
The Backlog is the Argument
Management's response to the numbers is the backlog. CoreWeave ended the quarter with approximately $104 billion in contracted revenue, and the company stated that this figure does not include more than $25 billion in new customer commitments secured in the first weeks of the third quarter. The magnitude comes, among other sources, from a $21 billion contract with Meta for the provision of AI cloud capacity through 2032, stacked on top of a previous commitment of $14 billion with the same client.
Translating into language a CFO can defend in an investment committee: each dollar of revenue recognized in the second quarter corresponds to more than $40 of contracted revenue that has not yet been delivered. If this ratio is indeed realized within the expected timelines, the viability of current investments is secured. If not, the leverage currently financing the capex becomes a structural problem at a time when refinancing debt is more expensive than the initial round of fundraising.
What This Evening Revealed About the Model
The CoreWeave case has become, in 2026, the main public test of the thesis that manufacturing AI capacity at scale is worth the required cost of capital. A quarter with $9.4 billion in capex, $640 million in interest, and $626 million in losses can only be sustained with a pipeline as described. Detractors present a consistent argument: infrastructure analysts from major banks have been warning since mid-year that the average duration of contracts, currently in the range of five to seven years, may be too short for the lifespan of a data center that needs to amortize equipment over eight to ten years, risking leaving excess capacity as the pricing curve for inference continues to decline.
The timing aligns with the previous day's announcement regarding the $500 billion platform that Nvidia forged with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. It is exactly this type of player, frontier lab, or next-generation cloud, that can now utilize GPUs financed by institutional vehicles rather than continue to weigh down the balance sheet with corporate debt. If the platform takes off as promised, CoreWeave's next earnings season may show a shift from interest rates to lines of capacity rental, indirectly impacting operational margins.
The Reading in Other Geographies
In the United States, the stock's behavior post-results dictates the company's own borrowing cost and directly influences competitive pricing among direct rivals, including Nebius, Crusoe, and Applied Digital. A positive reaction could reopen the long-term debt issuance window that has been narrower since the second quarter.
In the UK and Germany, where CoreWeave announced data center expansions in 2025 and 2026 respectively, the balance sheet is also read as an indicator of sovereign supply risk: the more the company depends on financial costs in dollars, the greater the exposure to pass-through pricing in euros and pounds, affecting inference prices for local clients in AI-intensive sectors such as financial services, chemicals, and telecommunications.
For a CIO needing to negotiate a GPU capacity contract for 2027, the number worth more than revenue or loss is the backlog. A supplier with approximately $129 billion contracted through 2032, much of which is already committed to Meta, has structural negotiating leverage over latecomers. The window to lock in reasonable pricing is closing faster than the new capacity pipeline can open.