CoreWeave is a cloud built specifically for AI workloads, offering GPU and CPU compute, storage, networking, managed services and bare metal servers on NVIDIA hardware. On-demand pricing is listed per hour and varies by GPU: roughly $6.50 for a single GH200, $42.00 for a GB200 NVL72 node, $49.24 for an eight-GPU HGX H100 node and $68.80 for an eight-GPU HGX B200 node, with discounted Spot rates on several models and per-GPU inference pricing. Newer options such as GB300 NVL72 are quoted through sales. The company also sells CoreWeave Forge, a layer for connecting the loop of running, observing and improving AI systems.
It is aimed at AI labs, foundation model builders, machine learning platform teams and enterprises that need dependable capacity for large training runs and production inference. CoreWeave states that 9 of 10 leading foundation model providers use it, which tells you the target scale. Smaller teams can start with single-GPU instances, while organizations needing dedicated hardware can move to bare metal. VFX and rendering studios are also named as a solution area.
Use cases include training large models on multi-rack NVIDIA clusters, serving inference at scale with consistent performance, and rendering visual effects. Operations teams use Mission Control to see every layer from GPUs to workload behavior, and the ARIA agent analyzes thousands of training runs to explain what changed and suggest the next iteration. Reliability features such as the Fleet Lifecycle Controller and Node Lifecycle Controller aim to recover from failures before they cost a run. Support is described as direct-to-expert, with a human answering first and no non-technical layer in between.
