Provider pricing

CoreWeave AI API pricing

CoreWeave is the largest specialized AI cloud, running 250,000+ NVIDIA GPUs including H100, H200, GB200 NVL72, and Blackwell Ultra. Pricing is per-GPU-hour with reserved-capacity discounts up to 60%, plus a managed Inference Service for serverless LLM endpoints with autoscaling and tensor-parallel sharding built in.

Founded2017
HeadquartersRoseland, NJ
Tracked modelsDirectory

Best for

Frontier labs and inference platforms that need thousands of GPUs on a tight schedule.

Strengths

  • Largest dedicated NVIDIA Blackwell and GB200 NVL72 fleet outside hyperscalers
  • InfiniBand fabric — 3.2 Tbps inter-node bandwidth for large training
  • CoreWeave Inference Service: serverless vLLM, SGLang, TensorRT-LLM
  • Reserved contracts under $2/H100-hour for multi-year commitments