11 providers · per GPU-hour
NVIDIA A100 cloud pricing
12 offers from 11 providers, 12 with a published on-demand price. Per GPU-hour, US$. Includes A100 40GB and A100 80GB. Prices checked 28 Sept 2026.
- Lowest on-demand
- $1.49/GPU-hr
- Lowest spot
- $1.21/GPU-hr
- Memory
- 40 GB HBM2
Spec sheet: A100 40GB ↗A100 80GB ↗
| # | Provider | Offer | On-demand | Spot | Committed | As published (original currency) · source | Actions |
|---|---|---|---|---|---|---|---|
| 1 | Jarvislabs | A100 x1 1× A100 80GB · 16 vCPU · 112 GB RAM | $1.49/GPU-hr | — | — | ||
| 2 | Hyperstack | A100 x1 1× A100 80GB | $1.60/GPU-hr | — | — | ||
| 3 | Verda (formerly DataCrunch) | A100 x1 1× A100 80GB | $1.73/GPU-hr | — | — | ||
| 4 | Krutrim Cloud | A100 x1 1× A100 80GB · 24 vCPU · 96 GB RAM | $1.99/GPU-hr | — | $1.03/GPU-hr 1-yr reserved | ||
| 5 | Crusoe | A100 80GB PCIe 1× A100 80GB | $2.00/GPU-hr | — | — | ||
| 6 | Cyfuture | A100 x1 1× A100 80GB · 8 vCPU · 64 GB RAM | $2.09/GPU-hr | — | — | ||
| 7 | E2E Networks | A100 x1 1× A100 80GB · 16 vCPU · 115 GB RAM | $2.10/GPU-hr | — | — | ||
| 8 | Crusoe | A100 x1 1× A100 80GB | $2.30/GPU-hr | — | — | ||
| 9 | OVHcloud | A100 x1 1× A100 80GB · 15 vCPU · 180 GB RAM · 300 GB disk | $2.66/GPU-hr | — | — | ||
| 10 | CoreWeave | NVIDIA A100 (8x) 8× A100 80GB · 128 vCPU · 2048 GB RAM | $2.70/GPU-hr | $1.21/GPU-hr | — | ||
| 11 | Lambda | A100 A100 80GB · 240 vCPU | $2.79/GPU-hr | — | — | ||
| 12 | Vultr | A100 | $2.80/GPU-hr | — | — |
What is the A100 used for?
The A100 is a data-centre GPU positioned by its maker for large-scale model training, built on the Ampere architecture and released in 2020. Multi-GPU servers link cards over NVLink 600 GB/s (SXM; bridge for 2 PCIe GPUs), which matters when a job is split across several GPUs.
How large a model fits on one A100?
With 40–80 GB of memory, one card holds the weights of a model up to about 20–40 billion parameters at 16-bit precision, or about 40–80 billion at 8-bit. That is weights only: inference also needs room for the KV cache, and training needs several times more for gradients and optimiser state, so larger jobs are split across multiple GPUs.
| Spec (as published) | A100 40GB | A100 80GB |
|---|---|---|
| Architecture | Ampere | Ampere |
| Form factor | SXM | PCIe | SXM | PCIe |
| Memory | 40 GB HBM2 | 80 GB HBM2e |
| Memory bandwidth | 1,555 GB/s | 2,039 GB/s |
| Interconnect | NVLink 600 GB/s (SXM; bridge for 2 PCIe GPUs) | NVLink 600 GB/s (SXM; bridge for 2 PCIe GPUs) |
| Max power (TDP) | 400 W | 400 W |
| Precisions listed | FP64, TF32, FP16, INT8 | FP64, TF32, FP16, INT8 |
| Partitioning | Up to 7 MIGs @ 5GB | Up to 7 MIGs @ 10GB |
| Released | 2020 | 2020 |
| Source | maker spec sheet ↗ | maker spec sheet ↗ |
Need A100 capacity for a longer run?
Committed and cluster pricing is often negotiated. Tell us the GPU count, duration and region.