As of November 4, 2026, the cheapest listed rate for the NVIDIA RTX 5090 is $0.25 per GPU-hour on Salad (community, Lowest priority). The cheapest on-demand rate is $0.50 on OpenRelay. Across 23 providers, the median rate card — each provider’s cheapest on-demand rate, community capacity and marketplace floors out — is $0.79 per GPU-hour, and the middle half of those rate cards sits between $0.66 and $1.19. Between rate cards, on-demand rates for the same card run from $0.50 (OpenRelay) to $1.60 (EmpirioLabs), 3.2× apart across 23 providers. By market, the cheapest listed rates are $0.41 spot (Meshive), $0.25 on community capacity (Salad), $0.50 on-demand (OpenRelay) and $0.59 reserved with no stated term (CloudRift). To buy, the median new-condition listing is $5,499.87 across 16 sellers. With 32 GB of VRAM, that is $0.016 per GB of memory per hour at the cheapest on-demand rate and $0.025 at the median.

The two cheapest on-demand rate cards for the NVIDIA RTX 5090, side by side on every card they both list: OpenRelay vs HexGrid.

Dated pieces on the NVIDIA RTX 5090: September 30, 2026 — Four rates for one RTX 5090, $0.50 down to $0.25, 30 September 2026. Each is frozen as of its date.

As of November 4, 2026, one provider prices the NVIDIA RTX 5090 on-demand as a whole 8-GPU node: $10.00 per node-hour (Limestone Networks, NVIDIA RTX 5090, dedicated 8-GPU node, $1.25 per GPU). 8 GPUs at the cheapest on-demand rate ($0.50 on OpenRelay) come to $4.00 an hour, where a provider rents 8 to one account and they need not share a unit. A unit price is what the provider quotes for the 8 GPUs together; whether it also rents one is not part of the quote. 1 prices it in units of 4, 1 prices it in units of 2 — the rate table below names each quote's own unit.

8 GPUs for a month, 24/7 (730 h) cost from $2,920 at the listed on-demand rate of OpenRelay ($0.50 per GPU-hour); the job bar above ranks every provider for a job of your own.