NVIDIA
A Hopper refresh that keeps H100 compute while substantially increasing memory capacity and bandwidth.
Memory-heavy inference, long-context LLM serving, HPC
| Provider | Price / hour | Region | Availability | Updated |
|---|---|---|---|---|
| CoreWeave | Not verified | US West | Contract / reserved | 28 Sept 2026, 17:19 |
| DigitalOcean | USD 4.47 | — | On-demand GPU Droplet; per-GPU hourly rate | 29 Sept 2026, 07:24 |
| Nebius | Not verified | Finland | On-demand | 28 Sept 2026, 17:19 |
| Nebius | USD 5.4 | — | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked | 7 Oct 2026, 06:04 |
| Runpod | USD 4.59 | — | Secure Cloud Pod; published on-demand rate | 29 Sept 2026, 07:24 |
| Verda | USD 4 | — | On-demand GPU instance; published rate | 29 Sept 2026, 07:28 |
Typically priced above H100 on the same provider.
A call centre consultancy leased four Nvidia H200 accelerators to assess whether running DeepSeek V4.1 Flash for Claude Code justified claims of substantial cost savings. The testing demonstrated that querying DeepSeek's hosted API directly was more economical than renting dedicated hardware.