NVIDIA H100 PCIe — what a GPU-hour costs
Listed rates, not availability. 11 rates read between and . · Renting for a year or more? Reserved quotes ↓
In September 2026 the H100 PCIe like-for-like median did not move.
Size your workload
Who is cheapest today, and what a month costs.
GPUsHours per monthRegionH100 PCIe by provider
One table per market, on-demand first; cheapest first within each.
| Provider · what is sold | $ / GPU-hour | 30 days | Read | Source |
|---|---|---|---|---|
 Beamfrom | $1.83−37% vs median | flat | captured | Deploy ↗ |
 GPU.aifrom | $1.98−31% vs median | flat · 22 d | captured | Deploy ↗ |
 Hyperstack | $2.50−13% vs median | flat | captured | Deploy ↗ |
 QuantaCloudMidwest - us-midwest-2 | $2.59−10% vs median | flat · 18 d | captured | Deploy ↗ |
 HexGridSecure Cloud | $2.75−5% vs median | flat · 28 d | captured | Deploy ↗ |
 Barrack AI | $2.890% vs median | flat · 10 d | captured | Deploy ↗ |
 RunPodAPISecure Cloud · the API request | $2.890% vs median | flat | captured | Deploy ↗ |
 OVHcloudAPIEU (FR) · €2.80 listed · ECB 1.1193 on October 23, 2026 | $3.13+8% vs median | flat | captured | Deploy ↗ |
 Scalewayfr-par-2 · €2.87 listed · ECB 1.1193 on October 23, 2026 | $3.21+11% vs median | flat | captured | Deploy ↗ |
| Scalewaypl-waw-2 · €2.87 listed · ECB 1.1193 on October 23, 2026 | $3.21+11% vs median | flat | captured | Deploy ↗ |
 Lambda | $3.29+14% vs median | flat | captured | Deploy ↗ |
 Cloud.ru Evolution549.00 RUB listed · Bank of Russia 0.012416 on October 24, 2026 | $6.82+136% vs median | flat · 25 d | captured | Deploy ↗ |
| Provider · what is sold | $ / GPU-hour | 30 days | Read | Source |
|---|---|---|---|---|
| Hyperstack−$0.50 (−20%) vs its on-demand $2.50 | $2.00−31% vs median | flat | captured | Deploy ↗ |
| Provider · what is sold | $ / GPU-hour | 30 days | Read | Source |
|---|---|---|---|---|
| RunPodAPICommunity Cloud · the API request | $1.99−31% vs median | flat | captured | Deploy ↗ |
 Vast.aifloorAPI$4.00/node ÷ 2 · Multi-GPU machines · 7d −9.5% | $2.00−31% vs median | flat | captured | Deploy ↗ |
 LiumfloorAPI | $2.05−29% vs median | thin marketfewer offers than a floor needs · last floor · not in the figures | Deploy ↗ | |
| GPU.aifrom | $2.40−17% vs median | stoppedlast confirmed · not in the figures | ||
| Vast.aifloorAPI | $2.77−4% vs median | thin marketfewer offers than a floor needs · last floor · not in the figures | Deploy ↗ |
| Provider · what is sold | $ / GPU-hour | 30 days | Read | Source |
|---|---|---|---|---|
| BeamGPU only | $3.50+21% vs median | flat | captured | Deploy ↗ |
| Provider · what is sold | $ / GPU-hour | 30 days | Read | Source |
|---|---|---|---|---|
| Hyperstackfrommin. commitment commit | $1.75−39% vs median | flat | captured | Reserve ↗ |
Other cards today, median rate card per GPU-hour: H100 SXM $3.53 (Price Index) · B200 $7.50 · H200 $4.54 · B300 $8.99 · GB300 NVL72 $18.00 · MI325X $2.25 · RTX 4090 $0.49 · all cards
Renting the H100 PCIe for a year or more?
The cheapest committed H100 PCIe rate each provider publishes, by term. A commitment is paid whether the GPU runs or not.
| Term | Per GPU-hour | Provider |
|---|---|---|
| Other commitment commit | $1.75 read | Hyperstack |
Across every card, a 1-year rate runs a median −15% against the same provider’s on-demand rate (7 providers); a 3-year rate −30% (5 providers). Every committed rate, card by card.
Get quotes for the H100 PCIe. A person at GPU Economy reads your request and passes it to up to three providers that list a rate for the card, the cheapest listed first; the quotes come from them, directly to you.
We may be paid a fee if a request becomes a contract; who receives it follows the listed rates, never a fee, and it changes no published rate or the Price Index (how, privacy).
As of , the cheapest listed rate for the NVIDIA H100 PCIe is $1.75 per GPU-hour on Hyperstack (reserved). The cheapest on-demand rate is $1.83 on Beam (from). Across 11 providers, the median rate card — each provider’s cheapest on-demand rate, community capacity and marketplace floors out — is $2.89 per GPU-hour, and the middle half of those rate cards sits between $2.54 and $3.17. Between rate cards, on-demand rates for the same card run from $1.83 (Beam) to $6.82 (Cloud.ru Evolution), 3.7× apart across 11 providers. By market, the cheapest listed rates are $2.00 spot (Hyperstack), $1.99 on community capacity (RunPod), $1.83 on-demand (Beam, from) and $1.75 reserved with no stated term (Hyperstack, from). To buy, the median new-condition listing is $34,917.35 across 9 sellers. With 80 GB of VRAM, that is $0.023 per GB of memory per hour at the cheapest on-demand rate and $0.036 at the median.
The two cheapest on-demand rate cards for the NVIDIA H100 PCIe, side by side on every card they both list: Beam vs GPU.ai.
How it is computed
What we measure
Publicly listed rental prices for 79 GPU models (NVIDIA and AMD) at 99 providers — 1167 current quotes across seven markets — and the retail price of the cards.
Where it comes from
Each provider's own pricing page or API; retail from reseller pages. Every figure links back to its source.
How it is normalised
Node prices are divided by GPU count, per-second rates multiplied out, a price listed in another currency shown as listed, beside its USD figure at the dated official rate (ECB; the Bank of Russia for RUB and VND). A reservation published as a total for its term is divided by the hours in that term, 8,760 per year.
What is modelled, not measured
One figure: cost per million output tokens, a listed rate divided by the throughput NVIDIA published for that card, published with the run that produced it in /token-cost.json, not on this page. Risk is not a model — it is read off the market a quote is from.
What we never do
Never estimate a price, never convert a currency silently, never publish a number a reader cannot verify on the linked page.
1167 quotes · 99 providers. Seven files under CC BY 4.0; history is stored append-only and never rewritten.
Weekly history since 2025, with a like-for-like table per provider: NVIDIA H100 PCIe price history.
Retail listings
Median new-condition store price: $34,917.35 across 16 listings from 9 sellers, most recent October 19, 2026.
| Seller | Condition, kind | Price | Seen | Source |
|---|---|---|---|---|
| ServerOrbit.com | new store price | $22,087.50 | Listing | |
| ServerOrbit.com | new store price | $27,332.70 | Listing | |
| Dihuni | new store price | $29,400 | Listing | |
| GotoDirect | new store price | $29,578 | Listing | |
| Network Outlet | new store price | $29,900 | Listing | |
| Uvation | new store price | $31,499 | Listing | |
| Tech Atlantix | new store price | $33,348.85 | Listing | |
| HardDrivesDirect | new store price | $33,995.95 | Listing | |
| ServerSupply.com | new store price | $35,838.75 | Listing | |
| ServerOrbit.com | new store price | $37,102.35 | Listing | |
| ServerOrbit.com | new store price | $37,102.35 | Listing | |
| ServerOrbit.com | new store price | $37,102.35 | Listing | |
| ServerOrbit.com | new store price | $37,102.35 | Listing | |
| itcreations.com | new store price | $37,995 | Listing | |
| ServerSupply.com | new store price | $38,570 | Listing | |
| ServerSupply.com | new store price | $39,900 | Listing |
At the cheapest on-demand rate ($1.83/hour on Beam, from), that purchase price equals about 19,081 GPU-hours of rental — roughly 26 months of continuous use. The lower reserved rate ($1.75 on Hyperstack) is left out: that capacity can be reclaimed or moves with supply. Power, hosting, cooling and idle time are not counted.
Compare the NVIDIA H100 PCIe with: NVIDIA H100 SXM
Every rate on this page as data: /gpu/h100-pcie/prices.json — the same fields as prices.json, narrowed to this card. CC BY 4.0, no key, no rate limit. On your own page: embed this card’s cheapest rates (/embed/gpu/h100-pcie, an iframe, no script).
Data updated: