Data
GPU Price Tracker
GPU cloud price tracker: hourly rates for H100, H200, B200, GB200 and MI300X across providers and regions, with availability and update times.
Last updated 7 Oct 2026, 06:10
| Availability | |||||
|---|---|---|---|---|---|
| AMD Instinct MI300X | Azure | Not verified | East US | 28 Sept 2026, 17:19 | Quota based |
| AMD Instinct MI300X | DigitalOcean | USD 2.59/hr | — | 29 Sept 2026, 07:24 | On-demand GPU Droplet; per-GPU hourly rate |
| NVIDIA B200 | CoreWeave | Not verified | US East | 28 Sept 2026, 17:19 | Limited |
| NVIDIA B200 | Nebius | USD 8.5/hr | — | 7 Oct 2026, 06:04 | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked |
| NVIDIA B200 | Runpod | USD 6.79/hr | — | 29 Sept 2026, 07:24 | Secure Cloud Pod; published on-demand rate |
| NVIDIA B200 | Verda | USD 6.11/hr | — | 29 Sept 2026, 07:28 | On-demand GPU instance; published rate |
| NVIDIA B200 SXM6 | Lambda | USD 6.99/hr | — | 7 Oct 2026, 06:10 | 1-GPU on-demand list rate; tax extra; capacity not checked |
| NVIDIA B300 | Nebius | USD 9.5/hr | — | 7 Oct 2026, 06:04 | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked |
| NVIDIA B300 | Runpod | USD 7.89/hr | — | 29 Sept 2026, 07:24 | Secure Cloud Pod; published on-demand rate |
| NVIDIA B300 | Verda | USD 7.5/hr | — | 29 Sept 2026, 07:28 | On-demand GPU instance; published rate |
| NVIDIA GB300 | Verda | USD 8.62/hr | — | 29 Sept 2026, 07:28 | On-demand GPU instance; published rate |
| NVIDIA H100 | AWS | Not verified | us-east-1 | 28 Sept 2026, 17:19 | Capacity blocks |
| NVIDIA H100 | CoreWeave | Not verified | US East | 28 Sept 2026, 17:19 | Contract / reserved |
| NVIDIA H100 | DigitalOcean | USD 4.41/hr | — | 29 Sept 2026, 07:24 | On-demand GPU Droplet; per-GPU hourly rate |
| NVIDIA H100 PCIe | Lambda | USD 3.29/hr | — | 7 Oct 2026, 06:10 | 1-GPU on-demand list rate; tax extra; capacity not checked |
| NVIDIA H100 PCIe | Runpod | USD 2.89/hr | — | 29 Sept 2026, 07:24 | Secure Cloud Pod; published on-demand rate |
| NVIDIA H100 PCIe | Scaleway | EUR 2.73/hr | — | 29 Sept 2026, 07:43 | Published per-GPU hourly price |
| NVIDIA H100 SXM | Lambda | USD 4.29/hr | — | 7 Oct 2026, 06:10 | 1-GPU on-demand list rate; tax extra; capacity not checked |
| NVIDIA H100 SXM | Nebius | USD 4.5/hr | — | 7 Oct 2026, 06:04 | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked |
| NVIDIA H100 SXM | Runpod | USD 3.49/hr | — | 29 Sept 2026, 07:24 | Secure Cloud Pod; published on-demand rate |
| NVIDIA H100 SXM | Scaleway | EUR 6.61/hr | — | 29 Sept 2026, 07:43 | Published per-GPU hourly price for H100-SXM configuration |
| NVIDIA H100 SXM | Verda | USD 3.25/hr | — | 29 Sept 2026, 07:28 | On-demand GPU instance; published rate |
| NVIDIA H200 | CoreWeave | Not verified | US West | 28 Sept 2026, 17:19 | Contract / reserved |
| NVIDIA H200 | DigitalOcean | USD 4.47/hr | — | 29 Sept 2026, 07:24 | On-demand GPU Droplet; per-GPU hourly rate |
| NVIDIA H200 | Nebius | USD 5.4/hr | — | 7 Oct 2026, 06:04 | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked |
| NVIDIA H200 | Nebius | Not verified | Finland | 28 Sept 2026, 17:19 | On-demand |
| NVIDIA H200 | Runpod | USD 4.59/hr | — | 29 Sept 2026, 07:24 | Secure Cloud Pod; published on-demand rate |
| NVIDIA H200 | Verda | USD 4/hr | — | 29 Sept 2026, 07:28 | On-demand GPU instance; published rate |
Rates marked “Not verified” are awaiting a confirmed provider or API source. We publish a price only once it can be attributed to the provider's own listing, and every row carries its own last-updated timestamp.
Sources attributed Data updated regularly Methodology
How GPU cloud pricing works
Cloud GPUs are usually billed per GPU-hour. On-demand rates are the most flexible and the most expensive; reserved or committed contracts, typically one to three years, trade flexibility for a lower hourly price. Spot or interruptible capacity is cheaper still but can be reclaimed at short notice.
What moves the price
The accelerator generation matters most: NVIDIA H100 and H200 instances remain the workhorses for training and inference, while Blackwell-generation B200 and GB200 systems command a premium where available. Interconnect (NVLink and InfiniBand), memory capacity, region, minimum cluster size and included storage or networking all change the effective cost. Always compare like-for-like configurations.
Prices in this table are gathered from public provider pages and marked with the date they were last checked. Where a figure cannot be verified it is shown as not verified rather than estimated.