Cheapest on-demand GPU rental, by provider type
For every GPU we track, the lowest publicly listed on-demand rate from each kind of provider: the 4 hyperscalers (AWS, Azure, Google Cloud, Oracle), 15 specialist GPU clouds, and the marketplace (Vast.ai). Prices only — no ranking by opinion. A tag names the tier or scope; “GPU only” means CPU and memory are billed separately, “floor” means a marketplace offer that moves with supply.
As of , the cheapest listed on-demand rates are:
- NVIDIA GB300 NVL72: $18.00 on Oracle Cloud (OCI) among hyperscalers; $8.62 on DataCrunch among specialist GPU clouds.
- NVIDIA GB200 NVL72: $16.00 on Oracle Cloud (OCI) among hyperscalers; $8.00 on GMI Cloud (from) among specialist GPU clouds.
- NVIDIA B300: $15.00 on Oracle Cloud (OCI) among hyperscalers; $6.94 on RunPod (Community Cloud) among specialist GPU clouds.
- NVIDIA B200: $14.00 on Oracle Cloud (OCI) among hyperscalers; $4.00 on GMI Cloud (from) among specialist GPU clouds; $7.75 on Vast.ai (floor) among marketplaces.
- NVIDIA H200: $7.91 on Amazon Web Services (us-east-1) among hyperscalers; $2.60 on GMI Cloud (from) among specialist GPU clouds; $3.98 on Vast.ai (floor) among marketplaces.
- NVIDIA H100 NVL: $2.59 on RunPod (Community Cloud) among specialist GPU clouds; $2.69 on Vast.ai (floor) among marketplaces.
- NVIDIA GH200: $2.29 on Lambda among specialist GPU clouds.
- NVIDIA H100 PCIe: $1.99 on RunPod (Community Cloud) among specialist GPU clouds; $2.64 on Vast.ai (floor) among marketplaces.
- NVIDIA H100 SXM: $6.88 on Amazon Web Services (us-east-1) among hyperscalers; $1.79 on Cudo Compute (no-kristiansand-1, GPU only) among specialist GPU clouds; $2.42 on Vast.ai (floor) among marketplaces.
- NVIDIA RTX PRO 6000 Blackwell: $1.69 on RunPod (Community Cloud) among specialist GPU clouds.
- NVIDIA A100 40GB: $2.74 on Amazon Web Services (us-east-1) among hyperscalers; $0.89 on Jarvislabs among specialist GPU clouds.
- NVIDIA L40S: $1.86 on Amazon Web Services (us-east-1) among hyperscalers; $0.79 on RunPod (Community Cloud) among specialist GPU clouds; $0.80 on Vast.ai (floor) among marketplaces.
- NVIDIA A100 80GB: $3.43 on Amazon Web Services (us-east-1) among hyperscalers; $1.19 on RunPod (Community Cloud) among specialist GPU clouds; $0.75 on Vast.ai (floor) among marketplaces.
- NVIDIA L40: $0.69 on RunPod (Community Cloud) among specialist GPU clouds.
- NVIDIA RTX 6000 Ada: $0.74 on RunPod (Community Cloud) among specialist GPU clouds; $0.61 on Vast.ai (floor) among marketplaces.
- NVIDIA RTX 5090: $0.69 on RunPod (Community Cloud) among specialist GPU clouds; $0.40 on Vast.ai (floor) among marketplaces.
- NVIDIA L4: $0.44 on Jarvislabs among specialist GPU clouds; $0.34 on Vast.ai (floor) among marketplaces.
- NVIDIA RTX A6000: $0.33 on RunPod (Community Cloud) among specialist GPU clouds; $0.47 on Vast.ai (floor) among marketplaces.
- NVIDIA RTX 4090: $0.34 on RunPod (Community Cloud) among specialist GPU clouds; $0.32 on Vast.ai (floor) among marketplaces.
- NVIDIA V100: $0.17 on DataCrunch among specialist GPU clouds.
Which card for which job
The table does not tell you which GPU to rent; the card decides that, not the provider. As a rule of thumb from the numbers above: training and large fine-tunes need the memory of an H100, H200 or B200; smaller fine-tunes and most inference of 7–70B models fit an A100 80GB,L40S or RTX 6000 Ada; small-model inference, image generation and hobby work run on an L4, RTX 4090or RTX 5090. Within a card, the cheapest listed rate is usually a marketplace floor or a community tier; the hyperscaler rate is what you pay for the same card with enterprise contracts and regions around it.
Cheapest on-demand rate per GPU-hour
On-demand list prices only: spot, reserved and negotiated rates are not compared here. Spot rates are on each GPU page; the weekly H100 series is thePrice Index.
Data updated: