Disclosure: I run gpuhourlyrates.com, and I used AI assistance to draft this post. Every figure below comes from the providers' own pricing pages on 15 November 2026. Prices change, so check the source before you rely on them.

Comparing GPU clouds by the hourly number alone is easy to get wrong. Here is the checklist I use.

1. Per GPU or per machine?

Some providers quote one GPU, others a whole instance. Even within one provider the per-GPU rate can depend on instance size. On Lambda's page, the H100 SXM is listed at $4.59 per GPU-hour in an 8x instance, $4.69 in 4x, $4.79 in 2x and $4.89 as a single GPU.

2. How finely is time billed?

RunPod states that Pods are billed per second. Hyperstack states that its pricing is billed per minute. A provider that rounds up to a full hour makes a ten-minute test cost a full hour.

3. Which tier is the price for?

  • Secure vs community. For the H100 SXM, RunPod shows $4.39 per hour on Secure Cloud and a Community Cloud rate starting from $2.69. The community figure is a "from" price and availability can differ.
  • On-demand vs spot. Hyperstack lists a spot price for the H100 PCIe of $2.00 per hour against $2.50 on-demand. Spot capacity can be interrupted.
  • On-demand vs reserved. Hyperstack reservation prices start from $2.72 for the H100 SXM against $3.20 on-demand, in exchange for a commitment.
  • Pods vs serverless. RunPod lists serverless H100 workers at $5.89 per hour against $4.39 for a Secure Cloud pod. Serverless can still be cheaper if a pod would otherwise sit idle.

4. What else is on the bill?

Storage is the usual surprise. RunPod lists container disk at $0.10 per GB per month, and volume disk at $0.10 per GB per month while running and $0.20 while idle. A 500 GB dataset on a $0.10 volume adds $50 per month before any GPU time. Check data transfer, public IPs, and any minimum top-up or credit expiry too.

5. Cost per job, not per hour

job cost = price per GPU-hour x number of GPUs x hours used + storage kept x storage price x months + idle time you were still billed for 

A training run on 8 GPUs for 6 hours at $3.00 per GPU-hour is 8 x 6 x $3.00 = $144. If the machine sits idle for two more hours before you stop it, that is another $48. Automatic shutdown often saves more than switching to a slightly cheaper provider.

6. Does the GPU fit your model?

A cheap GPU that runs out of memory saves nothing. Check VRAM before you compare prices.

I keep a daily-updated comparison read from the providers' own pages, with the date each price was checked, at https://gpuhourlyrates.com. Corrections are welcome in the comments.