Cloud GPU market splits across hyperscalers, neoclouds and marketplaces

The cloud GPU market is organized around three provider tiers: hyperscalers, specialist neoclouds and community marketplaces. The comparison tracks 67 cloud GPU providers and 14 curated GPU model families spanning more than 4,500 distinct instance configurations, with providers evaluated by catalog depth, billing tiers and availability of leading-edge chips such as B200, B300, MI300X and RTX 5090.

Hyperscalers such as AWS, Microsoft Azure, Google Cloud Platform and Oracle Cloud Infrastructure bundle GPU rental with broader cloud services, compliance programs and enterprise integration. They typically price 3-6x above specialist neoclouds for the same GPU. AWS spans 15 GPU model families, Azure spans 10 GPU model families and includes B200 and AMD MI300X, while GCP lists B200 through the A3 Ultra family and OCI spans 13 GPU model families with a bare-metal-first approach.

Neoclouds focus on GPU and AI infrastructure and typically undercut hyperscaler pricing by 50-80%. Lambda Labs, CoreWeave, RunPod, Crusoe, Nebius and others compete on GPU availability, startup speed, reserved capacity and deployment tooling. Community marketplaces such as Vast.ai, Salad and Theta EdgeCloud aggregate distributed inventory, often reaching the low end of public pricing but with more variable uptime, availability and hardware consistency.

Deployment models include serverless GPU, virtual GPU and bare-metal GPU. Serverless works best when utilization is under 30-40%, while provisioned or reserved capacity becomes cheaper for sustained workloads. Bare-metal options suit large training runs, HPC workloads and latency-sensitive jobs where virtualization overhead matters.