Cheapest A100 GPU Rentals
Last reviewed on 2026-04-30 — verify current rates with the provider before committing.
$1.65
Lowest A100 80GB/hr
$3.67
Highest A100 80GB/hr
15
Providers in this table
7
Regions covered
| Rank | Provider | GPU | $/Hour | $/Day | $/Month | Region | Availability | Action |
|---|
💡 General savings strategies
- Spot or preemptible capacity typically prices 30–70% below on-demand. See the spot instance guide for trade-offs.
- Marketplace providers (Vast.ai, FluidStack) usually list the lowest single-GPU rates; hyperscalers list the highest but offer the deepest features.
- For 8+ GPU clusters, contact providers directly — listed hourly rates rarely reflect committed-use discounts.
- Cross-region price differences for the same SKU can be material; always check at least two regions before committing.
- Reserved or 1-year commitments commonly save 30–50% versus on-demand for predictable workloads.
📊 What to look for in an A100 rental
An A100 listing is more than its hourly rate. The lines below explain how to read a provider's pricing page so you can compare like-for-like:
- Memory variant. 40GB and 80GB are different SKUs at different price points. The 80GB variant is what most large-model fine-tunes need.
- Single GPU vs. multi-GPU node. Some providers only sell 8x boxes; the per-GPU rate on those is often lower but the absolute monthly cost is much higher.
- Region. The same SKU in the same provider can differ by 10–20% across regions, and not every region keeps capacity on hand.
- Networking. NVLink and InfiniBand are usually included on hyperscaler 8x nodes but rarely on marketplace listings — important for distributed training, less important for inference or single-GPU work.
- Storage and egress. The hourly GPU rate excludes block storage and data egress, both of which can add 15–30% to the realized monthly cost.
For a workload-specific cost projection, plug your hours and GPU count into the cost calculator. For the cost categories that the GPU rate does not include — block storage, egress, networking — see hidden cloud GPU costs. For the discount structures available on top of the on-demand rate, see on-demand vs reserved vs spot.