Untangling the GPU-hour.

A list price is where it starts. Work out what your GPU-hour costs.

Check your quote Talk to us

GPU

H100 SXMH200

Same H100 chip, a

3.4× spread in list prices across 18 providers

H100 SXM on-demand · P90 ÷ P10 · n = 18October 20261

Same H200 chip, a

2.6× spread in list prices across 13 providers

H200 on-demand · P90 ÷ P10 · n = 13October 20264

Three H100 contract types, three median list prices

USD per GPU-hour, list priceOctober 20262

Spot / preemptiblen = 5$2.73
3-year+ reservedn = 5$2.97
On-demandn = 18$3.99

If unused hours are billed, a 25% reservation discount needs 75% use

Hypothetical: on-demand $1.00, reserved $0.75. $0.75 ÷ $1.00 = 75%.

Hypothetical prices, not market data3

Try yours.

Your price per GPU-hour ($) Share of hours you expect to use (%)

Committed: billed whether used or not

–

per GPU-hour you actually use

H100 SXM on-demand median $3.99, October 2026 · break-even at your price: –

Run the full check

Runs in your browser. Nothing you type is sent or saved.

Renting GPUs in the next few months? Tell us what's been hardest to compare.

Talk to us hello@siliqin.com

Notes

H100 SXM on-demand list prices per GPU-hour, 18 providers, collected in October 2026. Median $3.99/GPU-hr. Middle 80%: $3.10 to $10.41, a 3.4× spread (P90 ÷ P10). List prices, not negotiated deals. Without the 4 hyperscalers (n = 14 of the 18 H100 prices): median $3.95, middle 80% $3.04 to $5.52, a 1.8× spread (P90 ÷ P10). One price per provider, the median of its listed rates. Some are "starting at" or promo prices. 5 of the 18 are 8-GPU machine prices ÷ 8; included CPU, RAM and storage vary. Sources: provider pricing pages and price APIs; 2 via third-party price-list mirrors. Method
Ranges: spot / preemptible $1.99–$6.13, 3-year+ reserved $2.24–$5.40 (min–max); on-demand $3.10–$10.41 (P10–P90). H100 SXM, USD per GPU-hour, list prices collected in October 2026, not negotiated deals. One price per provider per row. Each row is a different set of providers: hyperscalers are 4 of the 18 on-demand prices but 3 of the 5 in each smaller row, so rows don't compare one for one. Spot prices move; these are snapshots. Many prices are 8-GPU machine prices ÷ 8, and 4 come via third-party price-list mirrors, including the 3-year+ median. Other reservation terms in our data aren't shown. Not shown: 1-year reserved (n = 4) and capacity block (n = 3), fewer than 5 prices each.
Hypothetical prices, chosen to show the arithmetic: not market data, a quote or an offer. If reserved hours are billed whether you use them or not, then at discount d, reserving costs less only above 1 − d utilization. Here d is 25%, so break-even is 75%. At 70% use the reservation costs $1.07 per hour used, $0.07 more than on-demand. Assumes no upfront payment, no resale, and on-demand capacity there when you need it. Education, not financial advice.
H200 on-demand list prices per GPU-hour, 13 providers, collected in October 2026. Median $5.40/GPU-hr. Middle 80%: $4.05 to $10.48, a 2.6× spread (P90 ÷ P10). List prices, not negotiated deals. Includes 4 hyperscalers; with only 9 other providers, they aren't shown separately. One price per provider, the median of its listed rates. SXM/HGX listings and listings that don't state the form factor; NVL and PCIe are left out, whether the listing or the provider's documentation says so. 5 of the 13 are 8-GPU machine prices ÷ 8; included CPU, RAM and storage vary. Sources: provider pricing pages and price APIs.