AWS lists 6 GPU models across 18 configurations, starting at $0.423 per GPU-hour for the L40S. It is the cheapest dedicated provider we track for 3 of its 12 GPU-and-pricing-model combinations: L40S (spot), H200 (spot), B300 (spot).
| GPU | VRAM | Pricing | AWS $/GPU-hr | Best dedicated | Difference | Cheapest at | |
|---|---|---|---|---|---|---|---|
| L40S | 48 GB | spot | $0.423 | $0.423 | cheapest | — | rent |
| H200 | 141 GB | spot | $1.85 | $1.85 | cheapest | — | rent |
| L40S | 48 GB | on-demand | $1.86 | $0.550 | +238% | Novita | rent |
| H100 SXM | 80 GB | spot | $1.92 | $0.970 | +98% | Google Cloud | rent |
| A100 SXM | 80 GB | spot | $2.15 | $0.410 | +425% | Google Cloud | rent |
| A100 SXM | 80 GB | on-demand | $3.43 | $1.09 | +215% | Thunder Compute | rent |
| B300 | 288 GB | spot | $3.69 | $3.69 | cheapest | — | rent |
| B200 | 192 GB | spot | $5.20 | $1.81 | +187% | Google Cloud | rent |
| H100 SXM | 80 GB | on-demand | $6.88 | $1.79 | +284% | Cudo Compute | rent |
| H200 | 141 GB | on-demand | $7.91 | $3.03 | +161% | Seeweb | rent |
| B200 | 192 GB | on-demand | $14.24 | $4.90 | +191% | GPU.ai | rent |
| B300 | 288 GB | on-demand | $17.80 | $6.60 | +170% | Massed Compute | rent |
How much does a GPU cost on AWS?
AWS starts at $0.423 per GPU-hour for the L40S on spot capacity. It lists 6 GPU models in total: A100 SXM, B200, B300, H100 SXM, H200, L40S.
Is AWS cheaper than other GPU clouds?
For 3 of the 12 GPU and pricing-model combinations it offers, AWS is the cheapest dedicated provider on this table — including L40S, H200, B300. For the rest, the table above names the provider that beats it and by how much. Comparisons are like-for-like on GPU, pricing model and capacity tier.
Which GPUs does AWS offer?
6 models: A100 SXM, B200, B300, H100 SXM, H200, L40S. Availability changes; the "in stock" flags on each GPU page reflect the last scrape.