Cloud GPU Providers
Compare GPUsCompare per-GPU-hour rental pricing across 29 clouds, from hyperscalers to GPU marketplaces. Rent B200s, H100s, A100s, RTX 4090s and more.
Missing a GPU cloud? Submit it for review.
Rental pricing by GPU
Vultr
Global cloud with GPU instances supporting both NVIDIA and AMD GPUs.
Hyperstack
AI cloud with on-demand, spot, and reserved GPU VMs. Per-minute prepaid billing.
Modal
Serverless GPU platform that runs Python functions, sandboxes and notebooks on autoscaling containers, billed per second with no idle instances.
Vast.ai
Peer-to-peer GPU marketplace with dynamic pricing. Per-second billing across 68+ GPU types.
Lambda Cloud
AI cloud with on-demand GPU instances and 1-Click Clusters for training and inference.
RunPod
GPU cloud with Community Cloud and Secure Cloud tiers. Per-millisecond billing with no minimum commitment.
Arvica Cloud
Australian GPU cloud (Arvica AI Holdings) brokering H100, H200, B200, B300, A100 and A40 capacity from partner data centres in Australia, APAC and elsewhere, 1-8 GPUs per instance, prepaid in USD; the region is confirmed at deployment.
Google Cloud
Google's hyperscale cloud with accelerator-optimized VMs (A4 B200, A3 H200/H100, A2 A100, G4 RTX PRO 6000, G2 L4) plus Spot VMs, CUDs and DWS Flex-start.
Crusoe
Climate-aligned GPU cloud powered by clean energy. Per-minute billing with no egress fees.
Nebius AI Cloud
AI-focused cloud with self-serve NVIDIA Blackwell and Hopper VMs in 1- or 8-GPU sizes, plus preemptible capacity and managed Kubernetes and Slurm.
CoreWeave
Specialized GPU cloud for AI/ML workloads. Multi-GPU instances with US and Europe availability.
Together AI
AI platform with dedicated GPU rental and GPU clusters alongside serverless inference.
Verda
European GPU cloud powered by renewable energy. Formerly DataCrunch. Pay-per-use pricing.
Runcrate
Runcrate is an AI cloud: rent GPUs on demand and call open-weight models through an OpenAI-compatible API. 20+ NVIDIA GPU types, from B300 to RTX A4000, run as VMs with SSH and Jupyter in 30+ cities across the US, Canada, Europe and Asia-Pacific. Prepaid, billed per second on most GPUs.
Baseten
Inference platform running custom and fine-tuned models on dedicated autoscaling GPUs with scale-to-zero, billed per minute with no idle charge; also sells per-token Model APIs.
Fluence
GPU marketplace renting 20+ GPU types, from RTX A4000 to B200, as VMs (1-8 GPUs) or Docker containers from partner data centers across North America, Europe, Asia and the Middle East. The first 2 hours are charged at launch; also runs GPU cluster auctions and cheap CPU VPS.
Oracle Cloud
Oracle's hyperscale cloud selling mostly bare-metal GPU shapes (GB300/GB200 NVL72, B300, B200, H200, MI355X, MI300X) at the same per-GPU price in every region.
AWS
Amazon's hyperscale cloud with EC2 GPU instances from T4 up to B300 (P6), plus Spot and Savings Plans, billed per second on Linux.