CoreWeave’s second-quarter revenue more than doubled year over year to $2.575B, but the neocloud’s spending grew just as fast. Operating expenses hit $2.624B, producing a $49M operating loss, and interest costs pushed the net loss to $626M.
The New Jersey company, one of the biggest rent-a-GPU businesses to emerge from the AI boom, is trying to move up the stack with managed inference services it expects to reach a $250M annual recurring revenue run rate by the end of 2026.
Growth is real but concentrated. Roughly 93 percent of the revenue increase came from existing customers, and three clients accounted for 72 percent of quarterly revenue, a dependency the company disclosed in its SEC filing. Total indebtedness stood at $35.6B as of June 30, with quarterly net interest expense up 140 percent to $640M. CoreWeave still has $10B available under revolving credit and delayed-draw facilities.
Capital spending is not slowing. The company expects 2026 capex of between $35B and $39B, betting that demand for AI compute has shifted from one-time training bursts to a continuous loop of training, inference, evaluation, and retraining.
CEO Michael Intrator told analysts that enterprises no longer treat a model as a deliverable, and that compute is becoming a recurring expense that grows with every application in production. The question investors face is whether that loop generates enough revenue to outrun a debt pile that keeps compounding.
