Below is a price snapshot for single A100 80GB GPU rental rates (as of May 2026). All numbers are on-demand, pay-by-the-hour in a U.S. region unless noted.
| Provider | Instance | On-Demand $/GPU-hr* | Notes |
|---|---|---|---|
| Thunder Compute | $0.78 | On-demand | |
| TensorDock | $0.85 | Marketplace-style pricing from public listings | |
| Paperspace | $3.18 | Public listed price | |
| Vast.ai | $1.21 | Average marketplace pricing. | |
| Sesterce | $1.24 | Public listed single-GPU equivalent | |
| FluidStack | $1.30 | Public estimate for on-demand access | |
| Verda | $1.29 | Public listed single-GPU equivalent | |
| Hyperstack | $1.35 | Public listed single-GPU equivalent | |
| Runpod Community Cloud | $1.39 | Community marketplace rate; varies by supply | |
| Crusoe Cloud | $1.65 | Public listed hourly pricing | |
| JarvisLabs | $1.49 | Pay-as-you-go A100 instance pricing | |
| AWS EC2 | p4de.24xlarge | $3.43 | $27.44/hr VM ÷ 8 GPUs |
| Lambda | $2.79 | Flat per-GPU rate | |
| Azure | ND24ads A100 v4 | $3.67 | Public listed single-GPU VM |
| Google Cloud | a2-ultragpu-1g | $6.09 | |
| Single-GPU costs are shown for multi-GPU nodes for easy comparison. Network, storage, and egress fees are not included. | |||
Methodology
(Why you can trust these numbers)
<ul><li><strong>On-demand only:</strong> No reserved-instance, commitment, or prepaid discounts.</li><li><strong>Same class of silicon:</strong> All providers offer NVIDIA A100 80 GB GPUs.</li><li><strong>Public price lists:</strong> Each figure comes from the vendor's current pricing page on the date above.</li><li><strong>USD in U.S. regions:</strong> Rates in other regions can differ by 5-20%.</li><li><strong>We update this data every month.</strong></li></ul>
Why this Matters for Developers
This chart compares how much it costs to run an NVIDIA A100 for 10 hours across major cloud providers.
Even small differences in hourly pricing add up quickly. A workload that costs $0.15 more per hour can translate into significantly higher monthly spend, especially for training jobs, batch processing, or production inference at scale.
Thunder Compute is ~7× cheaper than GCP for a single A100 and still 1.5-4.4× cheaper than AWS, Lambda, Paperspace, and Azure.
AWS P4 Instances
AWS offers A100 GPUs in its EC2 P4 Instances (P4d with A100 40GB and P4de with A100 80GB). All P4 instances have:
<ul><li>8 NVIDIA A100 GPUs</li><li>400 GB/s network performance</li><li>NVSwitch for 600 GB/s bidirectional throughput </li><li>8 x 1 TB NVMe SSD storage with 16 GB/sec read</li><li>96 vCPUs</li><li>1152GB RAM</li></ul>
As usual for AWS instances, egress fees are $0.09/GB after the first free 100GB.
| Region | p4d.24xlarge | p4de.24xlarge |
|---|---|---|
| us-east-1 | $21.95 | $27.45 |
| us-east-2 | $21.95 | N/A |
| us-west-1 | N/A | N/A |
| us-west-2 | $21.95 | $27.45 |
| ca-central-1 | $25.25 | N/A |
| ca-west-1 | N/A | N/A |
| Hourly pricing for US and Canada. Last update: May 11, 2026. | ||
Google Cloud A2 Instances
Through A2 instances, GCP offers accelerator-optimized compute with NVIDIA A100 GPUs:
<ul><li>A2 Standard (a2-highgpu and a2-mega) with A100 40GB.</li><li>A2 Ultra (a2-ultra) with A100 80GB.</li></ul>
| Instance name | CPUs | RAM | GPUs | SSD Storage | Price/hr |
|---|---|---|---|---|---|
| a2-ultragpu-1g | 12 | 170GB | 1 | 1x375GB | $6.09 |
| a2-ultragpu-2g | 24 | 340GB | 2 | 2x375GB | $12.19 |
| a2-ultragpu-4g | 48 | 680GB | 4 | 4x375GB | $24.37 |
| a2-ultragpu-8g | 96 | 1360GB | 8 | 8x375GB | $48.74 |
Final Thoughts on NVIDIA A100 Pricing
The NVIDIA A100 is still a widely used accelerator for AI training and inference, but pricing varies greatly across providers.
Hyperscalers like AWS, Azure, and Google Cloud offer mature ecosystems. While smaller neoclouds charge lower hourly costs for the same hardware.
If you are also considering renting the NVIDIA H100, check out NVIDIA A100 vs H100 and NVIDIA H100 Pricing.