Skip to content
  1. AI index
  2. Compute

AI compute, priced.

32 GPU offers across the Akash open market and the AWS and Azure price sheets, compared per GPU-hour and per GB of memory. Akash prices come from its public order book, and each market shows the date it was read.

H100 on the open market$2.56Median per GPU-hour, Akash
Price floor$0.14RTX 3060TI 8GB
Most memory for rent141 GBH200 141GB, $4.45/hr
Free on Akash now101 / 418GPU units not rented

What a GPU-hour costs

Price per GPU-hour against memory per GPU, for every offer on the Akash open market and the AWS and Azure price sheets. Both axes are log scale. Akash is the median asking price; the clouds are list price divided across the GPUs in the instance.

  • Akash
  • AWS
  • Azure

The cheapest GPU with 80 GB or more: $1.84 on Akash, $3.43 on AWS and $3.67 on Azure. Same memory class, 2.0× apart.

Akash Sep 30, 2026, hourly · AWS Sep 25, 2026 · Azure Jul 27, 2026All GPU boards

H100 rental price

Median asking price for an H100 80 GB on Akash, each point one published snapshot.

GPU price index

Median price across every GPU configuration listed on Akash, per snapshot.

Cheapest memory for large models

Price per GPU-hour divided by the GPU's memory, for every GPU with 40 GB or more: the cost of holding a large model in memory for an hour.

  • Akash
  • AWS
  • Azure
  1. 1PRO 6000SE 96GBAkash · 96 GB$0.0213
  2. 2A100 80GBAkash · 80 GB$0.0230
  3. 3H200 141GBAkash · 141 GB$0.0316
  4. 4H100 80GBAkash · 80 GB$0.0320
  5. 5g6e.xlargeAWS · 48 GB$0.0388
  6. 6p4de.24xlargeAWS · 80 GB$0.0429
  7. 7NC24ads A100 v4Azure · 80 GB$0.0459
  8. 8ND96amsr A100 v4Azure · 80 GB$0.0512
  9. 9g6e.12xlargeAWS · 48 GB$0.0546
  10. 10p5en.48xlargeAWS · 141 GB$0.0561
  11. 11p4d.24xlargeAWS · 40 GB$0.0686
  12. 12NC40ads H100 v5Azure · 94 GB$0.0743
  13. 13NC80adis H100 v5Azure · 94 GB$0.0743
  14. 14ND96isr H200 v5Azure · 141 GB$0.0752
  15. 15p5.48xlargeAWS · 80 GB$0.0860
  16. 16p5.4xlargeAWS · 80 GB$0.0860
  17. 17ND96isr H100 v5Azure · 80 GB$0.1536
Hover a bar for the full record

Memory on the newest accelerators

High-bandwidth memory per accelerator, from first-party vendor specifications. Memory decides the largest model one chip can hold; bandwidth decides how fast it reads it.

  1. 1AMD Instinct MI325X6 TB/s bandwidth256 GB
  2. 2AMD Instinct MI300X5.3 TB/s bandwidth192 GB
  3. 3NVIDIA B200 SXM8 TB/s bandwidth180 GB
  4. 4NVIDIA H200 SXM4.8 TB/s bandwidth141 GB
  5. 5NVIDIA H100 SXM3.35 TB/s bandwidth80 GB

We rank the market. We can rank you.

The same index that prices every GPU-hour tracks which brands ChatGPT, Claude and Gemini recommend when buyers ask. Want us to rank yours?

Get my brand ranked →