
GPU specialist against peer marketplace, priced against each other on the 8 GPUs both of them rent. Published per-GPU on-demand rates as each provider last stated them, with what their own pages say about billing, commitment and egress beside the numbers.
Vast.ai is cheaper on 5 of the 8 cards both of them rent, by a median of 17%. GPU.ai is cheaper on the other 3.
These are not the same product. A GPU specialist and a peer marketplace differ in who your counterparty is, which is why the rates below can be far apart without either being wrong. The prices are comparable; what stands behind them is not.
The 8 cards both of them rent, cheapest published per-GPU on-demand rate each. Dearest first.
| GPU | GPU.ai | Vast.ai | Difference |
|---|---|---|---|
| NVIDIA H200 SXM5 141GB | $3.59/hr | $4.08/hr | 12% cheaper on GPU.ai |
| NVIDIA H200 NVL 141GB | $3.29/hr | $3.60/hr | 9% cheaper on GPU.ai |
| NVIDIA H100 NVL 94GB | $3.19/hr | $2.92/hr | 8% cheaper on Vast.ai |
| NVIDIA H100 SXM5 80GB | $3.49/hr | $1.87/hr | 46% cheaper on Vast.ai |
| NVIDIA H100 PCIe 80GB | $2.48/hr | $2.47/hr | 0% cheaper on Vast.ai |
| NVIDIA RTX PRO 6000 Blackwell 96GB | $1.99/hr | $1.01/hr | 49% cheaper on Vast.ai |
| NVIDIA RTX 6000 Ada 48GB | $0.65/hr | $0.54/hr | 17% cheaper on Vast.ai |
| NVIDIA RTX A6000 48GB | $0.40/hr | $0.40/hr | 1% cheaper on GPU.ai |
Per GPU per hour, on-demand list prices as each provider last published them. Rates exclude tax and vary by region; a cheaper rate is not the same machine, the same network or the same support.
| GPU.ai | Vast.ai | |
|---|---|---|
| Billing granularityOn short jobs this outweighs the rate: an hour minimum on a 90-second run costs more than a 30% dearer per-second rate. | Per second | Per second |
| Minimum commitment | $5 minimum deposit, no term | $5 minimum to start |
| Payment model | Prepaid credit balance with burn rate and runway | Prepaid credits; optional auto top-up |
| Egress | not stated | Priced per byte by each host |
| Regions | 8 regions: US East/Central/West, Canada, EU West, EU East, India South, Asia-Pacific | 40+ data centres worldwide |
Neither site tells you whether your data trains their models, an uptime SLA, whether you get credits when it breaks or whether a commitment auto-renews. We looked on both and record the absence rather than filling it in — if you need any of it, ask before you commit.
Vast.ai is cheaper on 5 of the 8 cards both of them rent, by a median of 17%. GPU.ai is cheaper on the other 3.
NVIDIA H200 SXM5 141GB, NVIDIA H200 NVL 141GB, NVIDIA H100 NVL 94GB, NVIDIA H100 SXM5 80GB, NVIDIA H100 PCIe 80GB, NVIDIA RTX PRO 6000 Blackwell 96GB, NVIDIA RTX 6000 Ada 48GB, NVIDIA RTX A6000 48GB — 8 in total, on published per-GPU on-demand rates.
Of the cards both rent, GPU.ai starts at $0.40 per GPU-hour and Vast.ai at $0.40.
No. GPU.ai is a GPU specialist — an operator whose whole business is renting these cards. Vast.ai is a peer marketplace — a stranger whose machine the platform vets to some degree. Reliability, bandwidth and how long the machine stays online vary by host, not just by price. The rates are comparable; what stands behind them is not.