11 GPUs priced today across 43 configurations in 1 region. On average 99.9% above the market median.
Prices collected 2026-09-18 · seen since 2026-09-11
Visit Amazon Web Services →Amazon Web Services's on-demand median per GPU-hour, per snapshot — its eight most expensive GPUs. Gaps are runs where the GPU was not listed.
Per GPU-hour, on-demand, 2026-09-18. vs market compares Amazon Web Services's own median for that GPU with the index median across every provider. Click a GPU for every provider's price.
| GPU | Amazon Web Services | Market | vs market | Per month | Sizes | Billing |
|---|---|---|---|---|---|---|
| $14.24 | $6.59 | +116% | $10,254 | 8x · 1 config | on-demand | |
| $7.91 | $4.45 | +78% | $5,697 | 8x · 1 config | on-demand | |
| $6.88 | $3.50 | +97% | $4,954 | 1x–8x · 2 configs | on-demand | |
| $3.90 | $2.66 | +47% | $2,809 | 8x · 1 config | on-demand | |
| $3.43 | $1.79 | +92% | $2,470 | 8x · 1 config | on-demand | |
| $3.39 | $1.47 | +130% | $2,437 | 1x–8x · 8 configs | on-demand | |
| $3.06 | $0.79 | +287% | $2,203 | 1x–8x · 3 configs | on-demand | |
| $2.74 | $1.48 | +85% | $1,976 | 8x · 1 config | on-demand | |
| $1.83 | $1.83 | 0% | $1,318 | 1x–8x · 8 configs | on-demand | |
| $1.60 | $0.80 | +101% | $1,155 | 1x–8x · 10 configs | on-demand | |
| $0.98 | $0.59 | +66% | $704 | 1x–8x · 7 configs | on-demand |
Regions seen: us-east-1.
Amazon Web Services has been in the index since 2026-09-11. The chart at the top draws its on-demand median per GPU on every snapshot; a gap is a run where the GPU was not listed, never a guess. No price moved by more than 1% between consecutive snapshots so far.
As of 2026-09-18, Amazon Web Services's on-demand H100 SXM works out to $6.88 per GPU-hour, 97% above the market median of $3.50. That is about $4,954 a month per GPU at 720 hours.
Across the 11 GPUs it prices, Amazon Web Services averages 99.9% above the market median, and is cheaper than the median on 0 of them. It is the single cheapest provider we track for A10G. As a hyperscaler its instance price bundles CPU, RAM and network, which is most of the premium.
Public on-demand price list for Linux, us-east-1. Instance price divided by GPU count. Prices are collected every six hours and every snapshot is kept, so the history above is what Amazon Web Services actually published on those dates. Nothing is estimated or interpolated.
No spot or interruptible rates were listed on the pricing we read. Reserved pricing, where it exists, is by quote and not published. Only on-demand prices enter the index median.