

GPU specialist against peer marketplace, priced against each other on the 4 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 4 of the 4 cards both of them rent, by a median of 35%.
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 4 cards both of them rent, cheapest published per-GPU on-demand rate each. Dearest first.
| GPU | Nebius AI Cloud | Vast.ai | Difference |
|---|---|---|---|
| NVIDIA B200 SXM6 180GB | $8.50/hr | $6.25/hr | 26% cheaper on Vast.ai |
| NVIDIA H200 SXM5 141GB | $5.40/hr | $4.08/hr | 24% cheaper on Vast.ai |
| NVIDIA H100 SXM5 80GB | $4.50/hr | $1.87/hr | 58% cheaper on Vast.ai |
| NVIDIA RTX PRO 6000 Blackwell 96GB | $1.80/hr | $1.01/hr | 44% cheaper on Vast.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.
| Nebius AI Cloud | 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 GPU-hour | Per second |
| Minimum commitment | $25 first payment; no term | $5 minimum to start |
| Payment model | Bank card or bank transfer; PDF invoices by email | Prepaid credits; optional auto top-up |
| Free credits | Promo codes grant free credits | not stated |
| Egress | Free | Priced per byte by each host |
| Uptime SLA | 99.50% for Compute; compensation 10–30% of fees | not stated |
| Regions | 6 public regions: Finland, France, Israel, Kansas City (US), UK ×2 | 40+ data centres worldwide |
Neither site tells you whether your data trains their models, whether a commitment auto-renews, how long you have to dispute a bill or how much warning you get before maintenance. 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 4 of the 4 cards both of them rent, by a median of 35%.
NVIDIA B200 SXM6 180GB, NVIDIA H200 SXM5 141GB, NVIDIA H100 SXM5 80GB, NVIDIA RTX PRO 6000 Blackwell 96GB — 4 in total, on published per-GPU on-demand rates.
Of the cards both rent, Nebius AI Cloud starts at $1.80 per GPU-hour and Vast.ai at $1.01.
No. Nebius AI Cloud 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.