CoreWeave rents GPU computing capacity to companies training and serving AI models, running its own data centres rather than reselling another provider's cloud. The offering is narrower than a general purpose cloud, which lets it tune networking, storage and scheduling for the way large training jobs actually behave. A small number of very large customers account for most of the revenue, and capacity is built ahead of demand using debt secured against the hardware itself. That combination of customer concentration and leverage makes contract renewals and the useful life of each accelerator generation central to the investment case.