CoreWeave Expands AI Stack as Inference Surges

CoreWeave is moving beyond GPU compute into networking, storage, and software as AI inference demand outpaces training workloads. Analysts Dave Vellante and John Furrier, speaking at the Fully Connected event, noted CoreWeave expects its training-to-inference ratio to shift from 50/50 to 10/90 within a year. Customers are increasingly seeking shorter contracts and spot pricing, which carry higher costs than long-term agreements. CoreWeave's new Forge software layer adds observability, security, and closed-loop evaluation. Analysts described AI infrastructure as a systems race where token output per watt, not just silicon availability, determines commercial advantage.