A100 price per hour, cloud by cloud
Published 27 October 2026 · prices on this page are live
On-demand, one A100 GPU-hour costs from about US$1.20 on GPU marketplaces to about US$5 on Google Cloud's 80 GB machines. The A100 comes with 40 GB or 80 GB of memory, and the 80 GB version costs more almost everywhere. The big clouds' prices below were checked on 27 October 2026; PowerPod's are live.
On-demand A100 prices, 27 October 2026
| Provider | Machine | GPUs | Per machine-hour | Per GPU-hour |
|---|---|---|---|---|
| AWS | p4d.24xlarge | 8 × A100 40 GB | US$21.96 | US$2.74 |
| AWS | p4de.24xlarge | 8 × A100 80 GB | US$27.45 | US$3.43 |
| Google Cloud | a2-highgpu-1g / -8g | 1 or 8 × A100 40 GB | US$3.67 / US$29.39 | US$3.67 |
| Google Cloud | a2-ultragpu-1g / -8g | 1 or 8 × A100 80 GB | US$5.07 / US$40.55 | US$5.07 |
| Azure | NC24ads A100 v4 | 1 × A100 80 GB PCIe | US$3.67 | US$3.67 |
| Azure | ND96asr v4 | 8 × A100 40 GB | US$27.20 | US$3.40 |
| Azure | ND96amsr A100 v4 | 8 × A100 80 GB | US$32.77 | US$4.10 |
| Lambda | 8× A100 SXM | 8 × A100 80 GB | US$22.32 | US$2.79 |
| RunPod | A100 SXM, Secure / Community Cloud | 1 × A100 80 GB | US$1.79 / US$1.39 | US$1.79 / US$1.39 |
| RunPod | A100 PCIe, Secure / Community Cloud | 1 × A100 80 GB | US$1.79 / US$1.19 | US$1.79 / US$1.19 |
On PowerPod right now
Per GPU, billed by the second, lowest price on a verified host today:
Why the prices are so far apart
- The big clouds sell whole machines. Most of their H100 and A100 offers are eight-GPU machines with large CPUs, memory and local disks included, priced for companies, with uptime commitments, support and compliance paperwork.
- Commitments cut their prices. One- to three-year reservations, Spot (which can be stopped at any time) and capacity booking cost far less than on-demand, if you can plan or tolerate interruptions.
- GPU clouds and marketplaces sell single GPUs. Lambda and RunPod's Secure Cloud run their own or partner data centres; RunPod's Community Cloud and PowerPod list vetted independent hosts, which is cheaper and comes with no uptime guarantee.
- What is not in the price differs. AWS, Google and Azure charge extra for storage and for data sent out; Lambda and RunPod say they charge no data-transfer fees; PowerPod bills disk per GB-month and network transfer per GB, at rates shown before you rent.
40 GB or 80 GB?
- 40 GB fits QLoRA fine-tuning of models up to about 30B parameters, Stable Diffusion and FLUX, and serving 13B models at full precision.
- 80 GB fits 70B-class QLoRA, longer sequences and bigger batches; it is the one to pick if you are unsure and the price gap is small.
- For many jobs an RTX 4090 or L40S is cheaper still: see which GPU do I need.
Sources
Questions
How much does an A100 cost per hour on AWS?
On demand in us-east-1, as of 27 October 2026: US$2.74 per GPU-hour on the p4d.24xlarge (eight A100 40 GB, US$21.96 an hour) and US$3.43 on the p4de.24xlarge (eight A100 80 GB, US$27.45 an hour).
How much does an A100 cost per hour on Google Cloud?
US$3.67 an hour for one A100 40 GB (a2-highgpu-1g) and US$5.07 for one A100 80 GB (a2-ultragpu-1g), on demand in us-central1, as of 27 October 2026.
How much does an A100 cost per hour on Azure?
US$3.67 an hour for one A100 80 GB on the NC24ads A100 v4 in East US; the eight-GPU ND96amsr A100 v4 is US$32.77 an hour, about US$4.10 per GPU, as of 27 October 2026.
What is the cheapest way to rent an A100?
By the hour on a GPU marketplace: from US$1.19 on RunPod’s Community Cloud on 27 October 2026, and from USD 0.65 an hour for an A100 SXM4 on PowerPod right now, billed by the second, with no uptime guarantee.