price of compute

NVIDIA H100 PCIe rental price

Median $2.60/GPU-hr on-demand · across 5 providers · Updated 2026-08-11 03:30 UTC · Anatomy ↓

Price history

$0.00$1.00$2.00$3.00$4.00May 12Oct 2Apr 1Jan 1May 5Aug 10live pipeline →$2.60$1.99$1.47

Each point is the daily cross-provider median for this SKU and pricing type (per-provider medians first, so no marketplace’s volume dominates). Gaps are gaps — we don’t interpolate missing data.

Dashed/hollow points before 2026-08-09: archival reconstruction from dated provider pricing-page snapshots (see methodology) — fewer providers, monthly resolution.

The spread: the same NVIDIA H100 PCIe ranges from $1.09 to $3.29 per hour across tracked providers right now — a 3.0× difference. This is why we exist.

Where to rent it

ProviderTypeRegion$/GPU-hrSeen
Thunder ComputeThunder ComputeOn-demand—$1.092026-08-11 03:29 UTC
Vast.aiVast.aiSpotAE$1.602026-08-10 15:15 UTC
RunPodRunPodCommunity—$1.992026-08-11 03:29 UTC
Vast.aiVast.aiOn-demandGB$2.002026-08-11 03:29 UTC
HyperstackHyperstackOn-demand—$2.602026-08-11 03:29 UTC
RunPodRunPodOn-demand—$2.892026-08-11 03:29 UTC
LambdaLambdaOn-demand—$3.292026-08-11 03:30 UTC

Prices are as listed by providers at scrape time; availability at a listed price isn’t guaranteed. Links marked ↗ may be affiliate links — they never affect prices shown or ordering.

Anatomy: why it costs this

Everything that has to exist before one H100 PCIe-hour can be rented — then the arithmetic on the parts that can be priced honestly.

The foundry. Nearly every datacenter GPU is fabbed here — the tightest bottleneck in the chain.TSMC4N process · packagingThe memory feeding this SKU — for HBM parts, the second great bottleneck.SK hynix · SamsungHBM2e · 80 GBDesign and the software moat — and most of the margin in the whole chain.NVIDIAGPU design & softwareHow this silicon becomes a rentable machine.Supermicro · Dell · Foxco…HGX/DGX & OEM serversFleets are financed; capital recovery is the largest slice of the hourly price.Capital marketsdebt & equityColo and hyperscale facilities; PUE lives here.Data centersland · shell · coolingGrid electricity including cooling overhead (PUE) — priced on the cost stack below.Power gridkWh per GPU-hourEvaporative-cooling upper bound — environmentally real, economically small.Waterliters per GPU-hourThe providers listing this SKU right now — the layer this site prices hourly.Clouds & marketplacesThunder Compute · Vast.ai · H…The number this site tracks.You$2.60/GPU-hr median
Electricity$0.05/hr1.8%0.54 kWh from the grid
Water (upper bound)$0.003/hr0.1%≤ 0.6 L, priced high on purpose
Capital · facility · network · staff · margin$2.55/hr98.1%everything the operating inputs don't explain

Power basis: vendor GPU TDP only — full-server draw runs meaningfully higher × PUE 1.54 × 8.71¢/kWh (0.54 kWh from the grid). Water bounded at 1.8 L/kWh and $5/m³ — deliberately high. The residual is hardware amortization, facility, staff, and margin — presented as a residual because vendors don’t publish server street prices, and we don’t print numbers we can’t stand behind.

Verified inputs & sources
Price alert

One email when the NVIDIA H100 PCIe daily median hits your number. No account, one-shot, the alert retires after it fires.

NVIDIA H100 PCIe, NVIDIA H100 SXM, and NVIDIA H100 NVL are tracked as separate SKUs — different interconnect, different performance, different market. Comparing them in one number would be wrong, so we don’t.

Last scrape: 2026-08-11 03:30 UTC · 1308 observations today across 16 providers · Failures logged, never interpolated.

Know what a GPU-hour costs before everyone else does.

One email a week: what moved, why, and where the spreads are. No sponsors choosing the content.

Free forever. Unsubscribe anytime.