Ask what an H100 costs to rent and the honest answer is "it depends who you ask." On 9 November we read the public on-demand price lists of eight GPU clouds. For the same chip in the same SXM form factor, the cheapest list price was $3.20 per GPU-hour at Hyperstack and the highest was $6.155 at CoreWeave. That's a 1.9× spread for something most buyers think of as a commodity.

Chart 1H100 SXM on-demand price per GPU-hour (USD)
  • Hyperstack$3.20
  • Together AI$3.99
  • RunPod$3.99
  • Nebius$4.50
  • Lambda$4.59
  • Verda (formerly DataCrunch)$4.987
  • CoreWeave$6.155
Sources and conditions
  • Hyperstack: 'NVIDIA H100 SXM' on-demand per-GPU hourly, billed per minute, prepaid balance. Page uses '$' and does not name a currency code. hyperstack.cloud · 2026-11-09
  • Together AI: GPU Clusters, on-demand pay-as-you-go; page says 'All prices are per GPU per hour'. Listed as 'NVIDIA HGX H100' (HGX = SXM baseboard). GPUs per instance not stated. Preemptible and reserved rates are lower. together.ai · 2026-11-09
  • RunPod: Per-GPU pod price, Secure Cloud. FAQ confirms Secure Cloud price; Community Cloud from $2.69/hr. runpod.io · 2026-11-09
  • Nebius: On-demand per GPU-hour, listed as 'NVIDIA HGX H100'. Excludes taxes including VAT. Long-term reservations are cheaper. nebius.com · 2026-11-09
  • Lambda: 'NVIDIA HGX H100, 80 GB (208 vCPU config)'. Page says these prices 'take effect November 16, 2026, at 00:00 UTC'; current price (as of capture) is only visible in a logged-in account. Price is per GPU per hour, plus sales tax/VAT/GST. GPUs per instance not mapped on the page. Smaller H100 SXM configs are $4.69-$4.89/GPU/hr. lambda.com · 2026-11-09
  • Verda (formerly DataCrunch): '1x H100 SXM5 80GB' on-demand GPU instance, pay-as-you-go. Page also has a EUR toggle; VAT treatment not stated. verda.com · 2026-11-09
  • CoreWeave: 'NVIDIA HGX H100' 8-GPU instance at $49.24/hr divided by 8. Europe lists the same on-demand price. Spot is cheaper. coreweave.com · 2026-11-09

List prices from each provider's official page, checked 2026-11-09. Excludes tax. Hover a bar for conditions.

These are list prices, and list prices are a starting point. Every provider whose page mentions it says committed or interruptible capacity is cheaper. Together says "preemptible and reserved rates are lower," Nebius says long-term reservations are cheaper, and CoreWeave lists cheaper spot capacity. So the chart shows what you pay if you turn up with a credit card and no commitment. That's how most teams start, and it's the number that sets their first budget.

The CoreWeave number needs a footnote

CoreWeave doesn't sell a single H100. Its H100 price is for an 8-GPU HGX instance at $49.24 an hour, which we divided by eight. If you need a whole node anyway, that's a fair comparison. If you wanted one GPU for a fine-tuning experiment, it isn't, because the smallest thing you can buy there costs about $49 an hour.

Lambda's figures need care too. Its page says its listed prices "take effect November 16, 2026," and the price before then is only visible inside a logged-in account. We show the new list price ($4.59 for its 208-vCPU H100 SXM configuration) because that's what new customers will pay from next week. It's a rise: PricingSaaS's Pulse tracker, which compared captures of the page from 2 and 9 November, records the H100 moving from $3.99 to $4.59 and the B200 from $6.69 to $7.69. We didn't see the old prices ourselves, so treat those as Pulse's figures. Smaller H100 SXM configurations on the same page run $4.69 to $4.89.

The H200 is sometimes the better buy

The more useful finding is how differently providers price the next chip up. The H200 carries 141 GB of memory against the H100's 80 GB, going by how Verda names the two instances. You'd expect that to cost a lot more. Sometimes it doesn't.

Chart 2H200 on-demand price per GPU-hour (USD)
  • Hyperstack · SXM$3.99
  • RunPod$5.29
  • Nebius · SXM$5.40
  • Together AI · SXM$5.99
  • OVHcloud$6.195
  • CoreWeave · SXM$6.305
  • Verda (formerly DataCrunch) · SXM$6.44
Sources and conditions
  • Hyperstack: 'NVIDIA H200 SXM' on-demand per-GPU hourly, billed per minute, prepaid balance. Page uses '$' and does not name a currency code. hyperstack.cloud · 2026-11-09
  • RunPod: Per-GPU pod price, Secure Cloud. FAQ confirms Secure Cloud price; Community Cloud from $3.59/hr. Variant not stated. runpod.io · 2026-11-09
  • Nebius: On-demand per GPU-hour, listed as 'NVIDIA HGX H200'. Excludes taxes including VAT. Long-term reservations are cheaper. nebius.com · 2026-11-09
  • Together AI: GPU Clusters, on-demand pay-as-you-go; page says 'All prices are per GPU per hour'. Listed as 'NVIDIA HGX H200' (HGX = SXM baseboard). GPUs per instance not stated. Preemptible and reserved rates are lower. together.ai · 2026-11-09
  • OVHcloud: 'From $49.56 /hour', footnoted '*Estimated pricing for 8 GPUs'; divided by 8. Starting price, estimated. Currency is the '$' on the EN page; VAT not stated. ovhcloud.com · 2026-11-09
  • CoreWeave: 'NVIDIA HGX H200' 8-GPU instance at $50.44/hr divided by 8. Europe lists the same on-demand price. Spot is cheaper. coreweave.com · 2026-11-09
  • Verda (formerly DataCrunch): '1x H200 SXM5 141GB' on-demand GPU instance, pay-as-you-go. Page also has a EUR toggle; VAT treatment not stated. verda.com · 2026-11-09

List prices from each provider's official page, checked 2026-11-09. Excludes tax. Hover a bar for conditions.

At CoreWeave, the 8-GPU H200 node is $50.44 an hour against $49.24 for the H100, a premium of about 2%. At Together it's 50% ($5.99 against $3.99). Hyperstack's H200 at $3.99 costs the same as an H100 at Together or RunPod.

For any workload that's limited by memory, such as long contexts, big batches or a model that only just fits, the useful measure is price per gigabyte of GPU memory, not price per GPU. On that measure CoreWeave's H200 is about $0.045 per GB-hour against $0.077 for its H100. The more expensive-looking chip is 42% cheaper for what you need.

Blackwell premiums range from 1.4× to 2×

The B200 has the narrowest price range of the three chips, from $6.00 at Hyperstack to $9.138 at Verda, a 1.5× spread.

Chart 3B200 on-demand price per GPU-hour (USD)
  • Hyperstack$6.00
  • Lambda · SXM$7.69
  • RunPod$7.99
  • Together AI · SXM$8.19
  • Nebius · SXM$8.50
  • CoreWeave · SXM$8.60
  • Verda (formerly DataCrunch) · SXM$9.138
Sources and conditions
  • Hyperstack: 'NVIDIA B200 (192GB)' on-demand per-GPU hourly, billed per minute, prepaid balance. Page uses '$' and does not name a currency code. hyperstack.cloud · 2026-11-09
  • Lambda: 'NVIDIA HGX B200, 180 GB'. Page says these prices 'take effect November 16, 2026, at 00:00 UTC'; current price (as of capture) is only visible in a logged-in account. Price is per GPU per hour, plus sales tax/VAT/GST. GPUs per instance not mapped on the page. lambda.com · 2026-11-09
  • RunPod: Per-GPU pod price, Secure Cloud. Secure Cloud tier inferred because the table price matches the page's structured data; not stated in visible text. runpod.io · 2026-11-09
  • Together AI: GPU Clusters, on-demand pay-as-you-go; page says 'All prices are per GPU per hour'. Listed as 'NVIDIA HGX B200' (HGX = SXM baseboard). GPUs per instance not stated. Preemptible and reserved rates are lower. together.ai · 2026-11-09
  • Nebius: On-demand per GPU-hour, listed as 'NVIDIA HGX B200'. Excludes taxes including VAT. Long-term reservations are cheaper. nebius.com · 2026-11-09
  • CoreWeave: 'NVIDIA HGX B200' 8-GPU instance at $68.80/hr divided by 8. Europe lists the same on-demand price. Spot is cheaper. coreweave.com · 2026-11-09
  • Verda (formerly DataCrunch): '1x B200 SXM6 180GB' on-demand GPU instance, pay-as-you-go. Page also has a EUR toggle; VAT treatment not stated. verda.com · 2026-11-09

List prices from each provider's official page, checked 2026-11-09. Excludes tax. Hover a bar for conditions.

Compare each provider's B200 with its own H100 SXM and the premium runs from 1.40× at CoreWeave to 2.05× at Together. RunPod (2.00×), Hyperstack (1.88×), Nebius (1.89×), Verda (1.83×) and Lambda (1.68×) fall between. A team that has standardized on one provider and wants to know whether moving to Blackwell pays off will get a very different answer depending on which provider that is.

What to do with this

  • Don't benchmark against a single provider's price. The spread is large enough that your choice of provider matters as much as your choice of chip.
  • Ask whether the price is per GPU or per node. Two of the eight providers we checked price H200s by the 8-GPU node, so the per-GPU figure is our division.
  • Price memory, not chips, if your workload is memory-bound.
  • Re-check next week. List prices move on fixed dates: Lambda's new prices take effect on 16 November. We'll re-read every page in this piece within 30 days, and any figure older than that gets flagged on the site.

All figures exclude tax and are in US dollars as listed. Hover over or tap any bar in the charts above for its exact conditions, or open "Sources and conditions" under each chart.