GPU Rental: Compare H100 Cloud Rates From Six Providers

On-demand list rates for the NVIDIA H100 80GB SXM run $3.99 to $12.29 per GPU-hour in October 2026, with a $6.52 median across the six providers GYGO compares: RunPod Secure Cloud and Lambda at $3.99, CoreWeave at $6.16, AWS at $6.88, Google Cloud at $11.06 and Azure at $12.29. Below the list rates sit RunPod Community Cloud at $2.69 on hosts without an enterprise SLA, OVHcloud quoted from $2.99, and spot or committed tiers. Marketplace providers such as Vast.ai publish no list rate because their hosts set prices. Rates were last read 13 October 2026.

RunPod Secure Cloud and Lambda both list the H100 80GB SXM at $3.99 per GPU-hour, Lambda on its 8-GPU instance (its 1x, 2x and 4x shapes list at $4.29). CoreWeave publishes a node rate of $49.24 per hour for 8 H100s, which is $6.16 per GPU, a 54% premium that buys Kubernetes-native orchestration and InfiniBand. On the A100 80GB the order changes: RunPod Secure Cloud $1.79, CoreWeave $2.70 and Lambda $2.79 per GPU-hour. All figures are on-demand list rates in US regions, read 13 October 2026.

For a single fine-tune the run time matters more than the hourly rate. At RunPod Secure Cloud's $3.99 an H100 finishes a Llama 70B QLoRA run in about 42 minutes for $2.79; the same run on its $1.79 A100 takes about 68 minutes and costs $2.03. A full Mistral 7B fine-tune runs about 2.1 hours on the H100 ($8.38) against 3.4 hours on the A100 ($6.09). Pick the H100 when engineer time is the cost that matters and the A100 when the GPU bill is. For multi-node runs, CoreWeave's InfiniBand fabric is the reason to pay its premium. The best GPU for LLM training guide walks through the throughput numbers.

RunPod Community Cloud lists the H100 80GB SXM at $2.69 per GPU-hour, 33% under its Secure Cloud rate, on hosts without an enterprise SLA. CoreWeave's interruptible H200 node lists at $20.93 per hour for 8 GPUs, $2.62 per GPU, if your job checkpoints well. Committed and reserved terms run below on-demand at every provider but are quoted rather than listed. Tell us the duration and we collect those quotes for you.

Rent while utilisation is uncertain. Buying an H100 at $36,125 per GPU (one eighth of a complete 8-GPU server, reseller-listed 13 October 2026) and colocating it pays back against renting at the $6.52 median list rate around month 11 at 80% utilisation; against RunPod Secure Cloud's $3.99 floor it takes until month 20. Most teams start on GYGO Rent and move to Buy plus colocation once a workload has held 70% utilisation for six months. The build-vs-rent guide runs the full three-year numbers.

H100 80GB SXM, H200 141GB, A100 80GB, GB200 and RTX 4090, from the providers above plus the hyperscalers. Published list rates in October 2026: H200 at $5.29 on RunPod Secure Cloud, $6.31 on CoreWeave and $7.91 on AWS per GPU-hour; GB200 at $10.50 per GPU on CoreWeave's 4-GPU NVL72 instance; RTX 4090 at $0.34 on RunPod Community Cloud. AMD MI300X has no on-demand rate published by the six compared providers, though others sell it. If you need a GPU or region not listed here, say so in the form and we quote it.

You submit the GPU model, count, duration and any CPU, memory or storage minimums in the form on this page. We match the requirement against providers with that capacity, collect comparable quotes, and send them to you within one business day. You contract directly with the provider you choose. GYGO charges you nothing; the provider pays us.