GPUs and compute
What an hour of AI compute costs on the open market and on cloud price sheets, which GPU clouds ChatGPT, Claude and Gemini recommend when asked, and who owns the accelerators. Prices and answers recorded by Rank.ai; ownership from Epoch AI.
Recorded Oct 30, 2026
$2.56an hour
is the median price to rent one NVIDIA H100 on the Akash open GPU market.
Source: the Akash public order book, recorded by Rank.ai in the H100 rental price chart.
Key findings
Every chart has its own link, embed code and PNG, and says whose data it is.
14 of 24answers recommending Runpod
Rank.ai asked the three engines GPU cloud buying questions and recorded every answer on Sep 17, 2026. 3 of the 16 clouds were never recommended.
A recommendation is an answer that names the cloud when asked which GPU cloud to use. Each answer can name several clouds.
6 vs 0CoreWeave, ChatGPT vs Claude
Claude never recommended CoreWeave, Google Cloud, Microsoft Azure or Hyperstack, each of which ChatGPT recommended at least once.
ChatGPT and Claude through their providers' APIs, asked the same GPU cloud questions on the same day.
4.8xAzure's highest H100 list price vs the Akash median
Akash median from its public order book on Oct 30, 2026; Azure retail prices normalised to one GPU, from its price sheet on Jul 27, 2026.
List prices before discounts and commitments. Azure VM names say which H100 variant each is.
5.45MH100e owned by Alphabet
Microsoft is second at 3.57M. Median estimates as of May 12, 2026, in NVIDIA H100 equivalents.
Epoch AI's median estimates; each carries an uncertainty band on the board.
Charts
Boards
16 ranked · hourly · as of Oct 30, 2026
| # | Name | Median Akash GPU price |
|---|---|---|
| 1 | NVIDIA P4 8 GB PCIe | $0.06 / GPU-hr |
| 2 | NVIDIA T4 16 GB PCIe | $0.13 / GPU-hr |
| 3 | NVIDIA RTX 3060TI 8 GB PCIe | $0.14 / GPU-hr |
| 4 | NVIDIA GTX1050 2 GB PCIe | $0.14 / GPU-hr |
| 5 | NVIDIA RTX 3090 24 GB PCIe | $0.15 / GPU-hr |
6 ranked · as of Jul 27, 2026
| # | Name | Normalized retail price per GPU |
|---|---|---|
| 1 | Azure NC24ads A100 v4 | $3.673 / GPU-hr |
| 2 | Azure ND96amsr A100 v4 | $4.096 / GPU-hr |
| 3 | Azure NC40ads H100 v5 | $6.980 / GPU-hr |
| 4 | Azure NC80adis H100 v5 | $6.980 / GPU-hr |
| 5 | Azure ND96isr H200 v5 | $10.600 / GPU-hr |
9 ranked · checked daily · as of Oct 29, 2026
| # | Name | Price per physical GPU |
|---|---|---|
| 1 | AWS EC2 g6.xlarge | $0.805 / GPU-hr |
| 2 | AWS EC2 g6.12xlarge | $1.150 / GPU-hr |
| 3 | AWS EC2 g6e.xlarge | $1.861 / GPU-hr |
| 4 | AWS EC2 g6e.12xlarge | $2.623 / GPU-hr |
| 5 | AWS EC2 p4d.24xlarge | $2.745 / GPU-hr |
6 ranked · as of May 12, 2026
| # | Name | Estimated owned compute (median) |
|---|---|---|
| 1 | Alphabet | 5,453,215 H100e |
| 2 | Microsoft | 3,574,563 H100e |
| 3 | Amazon | 2,508,034 H100e |
| 4 | Meta | 2,402,919 H100e |
| 5 | Oracle | 1,204,217 H100e |
5 ranked · as of Jul 26, 2026
| # | Name | Accelerator memory capacity |
|---|---|---|
| 1 | AMD Instinct MI325X | 256 GB |
| 2 | AMD Instinct MI300X | 192 GB |
| 3 | NVIDIA B200 SXM | 180 GB |
| 4 | NVIDIA H200 SXM | 141 GB |
| 5 | NVIDIA H100 SXM | 80 GB |
Accelerator profiles
Sources
Free to use with attribution. Download the prices and the recorded answers, or copy a citation.
Every active, positive-price listing on the public order book, reduced to a median per GPU-hour.
Retail VM prices, divided by the GPUs in each VM.
On-demand instance prices per physical GPU.
Which GPU cloud ChatGPT, Claude and Gemini recommend, 16 clouds.
Estimated AI compute owned, in H100 equivalents, with uncertainty bands.
Open market prices are medians of real listings, so one scarce card or one dumped listing cannot move them. Cloud prices are list prices before discounts and commitments, normalised to one GPU.
The recommendations come from the same GPU cloud buying questions asked of each engine through its API, with every answer kept. A cloud counts as recommended when an answer names it for the question asked.
Every GPU listing on the Akash, AWS and Azure boards: market, GPU, memory, price per GPU-hour, date.
Each cloud's recommendation count, overall and by engine, with the answers asked.
Last updated October 30, 2026
Enter your website. In about two minutes, Rank.ai asks ChatGPT, Claude and Gemini 12 questions your customers ask and grades how often they name you.
