CoreWeave reported adjusted EBITDA of $838 million for the third quarter of 2025, delivering an EBITDA margin above 60%. Management attributed the strong profitability to its AI-optimized infrastructure, which includes purpose-built GPU clusters that drive utilization above 90%. Nearly 75% of capacity is under multi-year contracts, providing revenue visibility through 2027. Capital expenditures rose by 45% year-over-year as the company expanded its GPU fleet, but high contracted demand has kept utilization near peak levels and supported the resilient margin profile.

Despite a 133% year-over-year increase in revenue to $1.36 billion, CoreWeave’s operating expenses surged to $1.31 billion in Q3, limiting operating margins to just 4%. Interest expense climbed to $310.6 million following a 89% increase in non-current debt, which stood at $10.3 billion at quarter end. Hardware depreciation and lease financing costs have intensified as the company rapidly scales its data center footprint. The net loss of $110 million, or $0.22 per share, underscores the challenge of balancing growth investment with path-to-profitability objectives in a capital-intensive industry.

CoreWeave is positioning itself to benefit from a potential surge in continuous inference workloads driven by agentic AI applications. The company currently operates 590 megawatts of active power capacity and has a contracted pipeline of 2.9 gigawatts expected online within the next 12 to 24 months. Industry forecasts project that agentic AI could expand the total addressable market for GPU compute toward $200 billion, growing at a 43% compound annual rate over the coming years. If these workloads materialize, CoreWeave’s focus on bare-metal clusters and high-density power deployments could create differentiated long-term demand and support further margin expansion.