GPU rental prices by week
This table shows the median on-demand price per GPU-hour across cloud providers for 32 GPU models, plus 3-, 6- and 12-month reservation prices where providers offer them. Week: 26 October - 1 November 2026. The data comes from GetDeploying’s open dataset (CC BY 4.0); we choose the models and compute the medians. You are free to cite the table or download it as a CSV file.
Source: GetDeploying (CC BY 4.0), median for the week to 1 November 2026 · Refreshed weekly.
| GPU | VRAM | Power | On demand, $ per GPU-hour | Reserved, median $ per GPU-hour | |||||
|---|---|---|---|---|---|---|---|---|---|
| Median | Lowest | Highest | Providers | 3 months | 6 months | 12 months | |||
| GB200 | 186 GB | 1,200 W | 16.00 | 10.50 | 27.04 | 3 | not offered | not offered | not offered |
| B300 | 288 GB | 1,100 W | 9.15 | 5.57 | 17.80 | 17 | 7.14 | 5.38 | 6.00 |
| B200 | 180 GB | 1,000 W | 7.99 | 3.75 | 14.24 | 22 | 6.78 | 6.13 | 8.44 |
| H200 | 141 GB | 700 W | 4.88 | 2.09 | 13.78 | 35 | 4.37 | 3.99 | 3.68 |
| H100 | 80 GB | 700 W | 3.83 | 1.80 | 12.29 | 39 | 3.42 | 3.09 | 3.21 |
| AMD MI325X | 256 GB | 1,000 W | 3.08 | 2.25 | 3.80 | 3 | not offered | not offered | 2.20 |
| AMD MI300X | 192 GB | 750 W | 2.92 | 2.59 | 7.86 | 6 | 3.10 | 2.37 | 2.24 |
| RTX PRO 6000 | 96 GB | 600 W | 2.34 | 0.65 | 8.29 | 43 | 1.61 | 2.10 | 2.35 |
| A100 | 80 GB | 300 W | 1.87 | 0.89 | 8.25 | 38 | 1.30 | 1.43 | 1.65 |
| L40S | 48 GB | 350 W | 1.57 | 0.38 | 7.58 | 33 | not offered | 1.03 | 1.09 |
| A10 | 24 GB | 150 W | 1.55 | 1.00 | 4.52 | 8 | not offered | not offered | 1.70 |
| P100 | 16 GB | 250 W | 1.51 | 1.28 | 2.22 | 4 | not offered | not offered | 1.04 |
| A5000 | 24 GB | 230 W | 1.38 | 0.25 | 1.85 | 5 | 0.40 | not offered | 0.35 |
| L40 | 48 GB | 300 W | 1.03 | 0.33 | 1.80 | 12 | not offered | not offered | not offered |
| RTX 6000 Ada | 48 GB | 300 W | 1.03 | 0.65 | 1.57 | 10 | not offered | not offered | not offered |
| V100 | 32 GB | 300 W | 0.97 | 0.14 | 3.90 | 17 | not offered | not offered | 1.43 |
| RTX PRO 4500 | 32 GB | 200 W | 0.92 | 0.72 | 4.09 | 5 | not offered | not offered | not offered |
| RTX 3080 Ti | 12 GB | 350 W | 0.85 | 0.16 | 0.92 | 3 | not offered | not offered | not offered |
| A40 | 48 GB | 300 W | 0.85 | 0.59 | 2.42 | 6 | not offered | not offered | not offered |
| L4 | 24 GB | 72 W | 0.83 | 0.40 | 3.40 | 14 | 0.41 | 0.56 | 0.66 |
| A4000 | 16 GB | 140 W | 0.76 | 0.17 | 1.04 | 5 | 0.30 | not offered | 0.27 |
| RTX 5090 | 32 GB | 575 W | 0.75 | 0.35 | 2.52 | 16 | 0.68 | not offered | not offered |
| RTX 4000 Ada | 20 GB | 130 W | 0.74 | 0.28 | 1.53 | 3 | not offered | not offered | not offered |
| T4 | 16 GB | 70 W | 0.69 | 0.20 | 4.35 | 10 | not offered | not offered | 0.58 |
| A30 | 24 GB | 165 W | 0.68 | 0.32 | 1.23 | 3 | not offered | 0.51 | 0.49 |
| RTX A6000 | 48 GB | 300 W | 0.59 | 0.35 | 5.09 | 16 | 0.65 | 0.89 | 0.70 |
| RTX 4090 | 24 GB | 450 W | 0.47 | 0.33 | 2.22 | 12 | 0.59 | not offered | 0.54 |
| RTX 5080 | 16 GB | 360 W | 0.30 | 0.26 | 0.39 | 3 | not offered | not offered | not offered |
| RTX 4080 | 16 GB | 320 W | 0.27 | 0.23 | 0.49 | 4 | not offered | not offered | not offered |
| RTX 3090 | 24 GB | 350 W | 0.23 | 0.10 | 0.50 | 7 | not offered | not offered | not offered |
| RTX 5060 Ti | 16 GB | 180 W | 0.14 | 0.12 | 0.18 | 3 | not offered | not offered | not offered |
| RTX 3070 | 8 GB | 220 W | 0.10 | 0.08 | 0.13 | 3 | not offered | not offered | not offered |
Rounded to the cent; the CSV has the unrounded medians. A dash means that fewer than two providers offered that reservation that week.
How the medians are made
- For each model we take the median across providers for the week: first each provider’s own median, then the median of those medians. A provider with many listings counts once, like every other provider.
- “Lowest” and “Highest” are single listings, not medians.
- These are advertised prices; they do not show that the capacity is actually available.
- The table lists models with at least three on-demand providers. A reservation price appears when at least two providers offer it.
- VRAM and power (TDP) come from GetDeploying’s GPU models file.
Our changes to the source data: a selection of models and terms, weekly medians, and added VRAM and power columns.
How to cite
Aura Digital, “GPU rental prices by week”, week of 26 October - 1 November 2026, https://auradigitalglobal.com/gpu-rental-prices/. Data: GetDeploying, GPU rental price history, CC BY 4.0.
You may reuse the table and the CSV under the same license, CC BY 4.0: name GetDeploying as the source of the data and link to this page.
What does a server earn at these prices?
The GPU ROI calculator turns these medians into payback time and annual return for your own server. The full arithmetic, electricity included, is in Renting vs Buying GPU Compute: the Real TCO in 2026.
Considering your own GPU server?
Tell us the model and the load. We reply within 24 hours, and a quote takes up to 3 working days.