Paid links may earn Compute Screener a referral fee. They never affect prices, weights or the order.
About NVIDIA H100
The NVIDIA H100 rental price today is $3.69 per GPU-hour, down 0.6% since yesterday, down 0.4% over the past week and up 3.9% over the past month. At that rate one H100 costs about $2,690 for a month of 730 hours. Over the past year it has ranged from $2.78 to $3.72. The index is the price of one benchmark contract: H100 SXM 80GB, on-demand, 8-GPU node, US hardware, CPU-neutral. It is standardized across 16 cloud providers and marketplaces, from Akash at $2.58 to CoreWeave at $6.16, and is updated every 2 to 3 hours.
The H100 is NVIDIA's Hopper-generation data-center GPU, announced on March 22, 2022 and built with 80 billion transistors on TSMC's 4N process. It succeeded the A100 and carried the 2023–2024 AI build-out: Meta trained Llama 3.1 405B on more than 16,000 H100s, and xAI's Colossus cluster opened with 100,000 Hopper GPUs. It is offered by more cloud providers than any other GPU on this site, and with the B200 it underlies the GPU rental futures that CME has filed and ICE has set for listing.
Specifications
| Spec | H100 SXM |
|---|
| Architecture | Hopper, TSMC 4N, 80 billion transistors |
| Memory | 80 GB HBM3, 3.35 TB/s |
| BF16 / FP16 tensor | about 990 TFLOPS dense, 1,979 with sparsity |
| FP8 tensor | about 1,979 TFLOPS dense, 3,958 with sparsity |
| NVLink | 900 GB/s |
| Power | up to 700 W |
| Other versions | H100 PCIe 80 GB at 2 TB/s; H100 NVL 94 GB at 3.9 TB/s |
| Typical server | HGX or DGX H100, 8 GPUs |
Source: NVIDIA H100, accessed October 24, 2026; PCIe figures from NVIDIA's H100 PCIe datasheet of October 2022.
Where the H100 is rented
Every large cloud rents H100s as 8-GPU machines. AWS p5.48xlarge became generally available on July 26, 2023 and Azure ND H100 v5 on August 7, 2023. General availability of Google Cloud A3 was announced for September 2023. Oracle rents BM.GPU.H100.8 machines. CoreWeave and Crusoe rent H100s as full 8-GPU nodes, while Lambda, Nebius and marketplaces such as Vast.ai and RunPod also rent single GPUs; GetDeploying tracked 57 H100 providers on October 23, 2026.
Terms run from interruptible spot capacity and hourly on-demand rental to reserved contracts of one to 36 months. Most rental value trades on contracts of six months or more, negotiated privately, and posted on-demand prices "are typically fixed and changed infrequently", SemiAnalysis wrote in April 2026.
What this price measures
The index is the posted on-demand price of one benchmark contract, an H100 SXM 80GB in an 8-GPU node of US hardware, rented by the hour from GPU clouds and marketplaces; prices for other form factors and node sizes are converted to that contract, and prices from other regions are counted without a regional adjustment for now. It is not a hyperscaler, reserved, spot or transaction price.
Those prices differ widely. On October 23, 2026, 8-GPU H100 SXM machines were listed at $3.90 at Crusoe, $3.99 at Lambda, $6.16 at CoreWeave, $10.00 at Oracle and $11.07 at Google Cloud per GPU-hour, while AWS showed $2.97 for a 36-month reservation. Crusoe and Lambda were both out of stock that day. Trackers that sample different providers, terms and dates publish very different H100 numbers; one comparison found an eightfold gap between a single marketplace spot listing in July 2026 and a hyperscaler on-demand benchmark for March and April, which its author notes are not like-for-like.
What moves H100 rental prices
Price history
Before October 23, 2026 this index's history is daily from November 7, 2025, rebuilt with this site's methodology from the posted prices recorded in SkyPilot's open price catalog for 5 to 8 providers a day, and weekly before that, from GetDeploying's H100 series; both are scaled to join the live index.
Futures and benchmarks
CME has filed H100 rental futures and ICE has set one for listing; neither was trading on October 24, 2026. CME Group's NYMEX contract GPU1 covers 730 GPU-hours, about a month, and settles on Silicon Data's H100 index; its planned October 5, 2026 launch slipped when the CFTC extended its review to November 9. ICE Futures U.S. has set its H100 contract, HPR, settled on the monthly average of Ornn's H100 index, for listing with no date.
The Bloomberg tickers beside the name are other H100 benchmarks: Silicon Data's SDH100RT, Ornn's ORNNH100, built from executed rentals, and SemiAnalysis's SAH100SC. Each measures a different market, so their levels differ from this index.
FAQ
How much does it cost to rent an H100 for a month? One H100 for 730 hours costs about 730 times the hourly price; the first paragraph above gives that figure at today's index.
Why are some H100s much cheaper? Many of the cheapest listings are PCIe cards, with 2 TB/s of memory bandwidth instead of 3.35 TB/s, single GPUs without fast networking, interruptible spot capacity or providers outside the US.
How does the H100 compare with the H200? The H200 has the same compute but 141 GB of faster memory, which matters most for serving large models. See the H200 page.
How often is this index updated? Every 2 to 3 hours, with one value per day in US Central time that is final when the day closes.
Sources
Text reviewed Oct 24, 2026; the figures in it are as of that date. The figures at the top of this page update every 2 to 3 hours. Not investment advice.
