joulemathBlog ·
What an H100 hour costs across 7 clouds (Oct 26, 2026)
How much is one H100 hour this week? We checked 7 public price lists and one trade index on October 26, 2026. Fixed on-demand prices run from $2.69 to $6.16 per GPU-hour. The index of real trades sits at $3.03.
Disclosure: this post has an affiliate link to Vast.ai. If you sign up through it, joulemath may earn a commission at no extra cost to you. It does not change any number here.
H100 SXM, per GPU per hour
| Where | Price | What it is |
|---|---|---|
| RunPod Community | $2.69 | 1 GPU, on-demand. Secure Cloud is $3.99. |
| Ornn OCPI-H100 | $3.03 | Index of executed rental trades, page updated Oct 25, 20:00 UTC |
| Massed Compute | $3.14 | 1x H100 SXM5, on-demand |
| Vast.ai median (affiliate) | $3.20 | Marketplace, 4 single-GPU offers: $1.93 low, $9.10 high |
| Together AI | $3.99 | GPU clusters, on-demand. $3.19 if reserved 181+ days. |
| Lambda | $4.29 | 1 GPU. $3.99 per GPU on an 8x instance. Plus tax. |
| Nebius | $4.50 | HGX H100, on-demand |
| CoreWeave | $6.16 | 8-GPU HGX H100 at $49.24/hr, divided by 8 (our math) |
All pages read October 26, 2026, about 09:00 to 10:00 CET. Vast numbers come from its public offer search (console.vast.ai/api/v0/bundles) at 09:06 CET: rentable, on-demand, exactly one H100 SXM. Four offers is a tiny sample, so the median can jump hour to hour. RunPod's page still says "Updated October 8, 2026".
What jumps out
- More than 2x spread. $2.69 at the low fixed price, $6.16 at the high one. Same chip, different packaging: single GPU vs 8-GPU boxes, support, network, contract terms.
- Commitment cuts the price. Together lists $3.99 on demand and $3.19 for 181+ days. Lambda's 8x box is $0.30 per GPU-hour cheaper than its single GPU.
- The trade index is in the middle. Ornn's $3.03 is up 6.32% over one month, per Ornn.
- Newer chips are moving faster. On the same Ornn page, H200 is $6.30 (+24.02% in a month) and B200 is $9.70 (+20.65%).
Per kW: what a host could gross
Our assumptions, not sourced: about 1,000 W per H100 including its share of the server, so 1 GPU per kW, rented 70% of the time.
| Rate per GPU-hour | Gross per kW per day |
|---|---|
| $2.69 (RunPod Community) | $45.19 |
| $3.03 (Ornn index) | $50.90 |
| $4.29 (Lambda 1x) | $72.07 |
| $6.16 (CoreWeave, per GPU) | $103.49 |
Math: rate x 24 hours x 70%. These are renter prices. A host on a marketplace keeps less after the platform's cut, and power at $0.06/kWh costs about $1.44 per kW per day. Hardware is the big cost: see the H100 page for buy price and payback.
Open the per-kW calculator
Last week's H200, A100 and RTX 4090 prices: GPU rental prices (Oct 21). How we calculate: methodology. Prices change constantly; this post won't. Nothing here is financial advice.