From $0.47 per GPU-hour at GPU.ai on 2026-09-18 — 32% under the $0.69 median across 9 providers.
Prices collected 2026-09-18 · every provider's own public pricing
Individual on-demand configurations, in stock, cheapest first — a listing is one instance type, SKU or marketplace offer, priced per GPU. Marketplace offers are single machines and can be gone by the time you look.
| # | Provider | Configuration | GPUs | $/GPU/hr | Month | vs median |
|---|---|---|---|---|---|---|
| 1 | 1x rtx-5090 · listed | 1 | $0.47 | $338 | −32% | |
| 2 | RTX 5090 ×1 · offer 47734276 · South Korea, KR | 1 | $0.48 | $347 | −30% | |
| 3 | RTX 5090 ×2 · offer 51495118 · Quebec, CA | 2 | $0.49 | $356 | −28% | |
| 4 | RTX 5090 ×1 · offer 51495121 · Quebec, CA | 1 | $0.50 | $357 | −28% | |
| 5 | 1x rtx-5090 · listed | 1 | $0.50 | $360 | −28% | |
| 6 | RTX 5090 ×1 · offer 49863106 · United States, US | 1 | $0.51 | $366 | −26% | |
| 7 | RTX 5090 ×1 · offer 51012197 · United States, US | 1 | $0.51 | $366 | −26% | |
| 8 | RTX 5090 ×2 · offer 51376107 · Zhejiang, CN | 2 | $0.52 | $375 | −25% | |
| 9 | RTX 5090 ×4 · offer 50595282 · California, US | 4 | $0.53 | $384 | −23% | |
| 10 | RTX 5090 ×2 · offer 50236334 · Hubei, CN | 2 | $0.53 | $385 | −23% |
Each provider reduced to its own median for the RTX 5090 — 4 of 9 sit below the market median.
| Provider | Kind | Median $/GPU/hr | vs market |
|---|---|---|---|
| marketplace | $0.50 | −28% | |
| marketplace | $0.68 | −2% | |
| marketplace | $0.68 | −1% | |
| marketplace | $0.68 | −1% | |
| marketplace | $0.69 | +0% | |
| marketplace | $0.73 | +6% | |
| marketplace | $0.89 | +29% | |
| neocloud | $0.91 | +31% | |
| neocloud | $0.99 | +43% |
Median of the providers headquartered in each country. Where the hardware sits can differ; this is who you contract with.
On 2026-09-18 the cheapest on-demand RTX 5090 we track is $0.47 per GPU-hour at GPU.ai (1x rtx-5090 · listed), about $338 a month per GPU. The market median across 9 providers is $0.69, so the cheapest listing is 32% below it.
Yes. The spot/interruptible median is $0.25 against $0.69 on-demand — 64% off — from 1 provider. The cheapest spot listing is $0.25 at SaladCloud. Spot can be taken away mid-job, which is why it never enters the headline median.
Because the market is three markets. Marketplaces are independent hosts competing on price with consumer-grade uptime; neoclouds sell the GPU more directly; hyperscalers bundle CPU, RAM, network and an ecosystem into the instance price. The median sits where most providers sit; the cheapest listing is usually a marketplace host or a promotional neocloud rate. 4 of 9 providers are below the median today.
Prices are re-read every six hours from each provider's own pricing and every snapshot is kept, so this page reflects the latest run and the history behind it. Recent cuts: Spheron -19% on 2026-09-18; Vast.ai -2.4% on 2026-09-18; Akash -1.5% on 2026-09-18.