Cheapest A100 rental per hour
The cheapest on-demand NVIDIA A100 rental in June 2026 is $0.58/hr (A100 40GB) on Vast.ai. The A100 is now one of the best-value datacenter GPUs: newer Hopper/Blackwell parts have pushed its price down sharply. The ranking below covers both the 40GB and 80GB variants across every provider that publishes a price.
Source: Provider pricing pages. Data as of June 2026.
A100 prices ranked cheapest first
| Rank | Provider | Variant | On-demand /hr | Spot/community /hr |
|---|---|---|---|---|
| #1 | Vast.ai | A100 40GB | $0.58/hr | — |
| #2 | Vast.ai | A100 80GB | $0.74/hr | $0.35/hr |
| #3 | RunPod | A100 40GB | $1.19/hr | — |
| #4 | Verda (formerly DataCrunch) | A100 40GB | $1.29/hr | — |
| #5 | Hyperstack | A100 80GB | $1.35/hr | $1.08/hr |
| #6 | RunPod | A100 80GB | $1.39/hr | — |
| #7 | Verda (formerly DataCrunch) | A100 80GB | $1.79/hr | — |
| #8 | Lambda | A100 40GB | $1.99/hr | — |
| #9 | Crusoe | A100 80GB | $2.00/hr | — |
| #10 | Vultr Cloud GPU | A100 80GB | $2.40/hr | — |
| #11 | CoreWeave | A100 80GB | $2.70/hr | — |
| #12 | AWS EC2 | A100 40GB | $2.75/hr | $1.42/hr |
| #13 | Lambda | A100 80GB | $2.79/hr | — |
| #14 | Paperspace (DigitalOcean) | A100 80GB | $3.18/hr | — |
| #15 | Microsoft Azure | A100 40GB | $3.40/hr | $0.75/hr |
| #16 | AWS EC2 | A100 80GB | $3.43/hr | $1.41/hr |
| #17 | Google Cloud | A100 40GB | $3.67/hr | $1.84/hr |
| #18 | Microsoft Azure | A100 80GB | $3.67/hr | $0.82/hr |
| #19 | Google Cloud | A100 80GB | $5.07/hr | $2.53/hr |
Source: Provider pricing pages. Data as of June 2026.
Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.
Frequently asked questions
What is the cheapest A100 rental per hour?
The cheapest on-demand A100 we track is $0.58/hr (A100 40GB) on Vast.ai. The A100 has become a strong value pick now that Hopper and Blackwell have pushed its price down. Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.
A100 40GB vs 80GB: which is cheaper and which do I need?
The 40GB is cheaper but limits the largest model/batch sizes. Pick 80GB if you fine-tune larger models or want bigger batches; pick 40GB for smaller models, research and tighter budgets. Both are widely available on neoclouds.
Is the A100 still worth renting in 2026?
Yes for cost-sensitive training, fine-tuning and inference of mid-sized models. The H100 is faster per dollar on the biggest jobs, but A100 prices have fallen enough that it is often the cheapest way to get 80GB of HBM. Compare both in the calculator.
Why is the A100 cheaper on neoclouds than on AWS/GCP/Azure?
Hyperscaler A100 instances bundle vCPU, RAM and support and (for the SXM variants) sell only as 8-GPU machines, so the derived per-GPU price is higher. Neoclouds sell the GPU with minimal overhead.
Keep exploring
Last updated: 2026-06-21