The cheapest usable H100 hour is $0.94 — and the same GPU can cost 12× more, before egress

Asked:

“What is the real effective price per hour for an NVIDIA H100 across GPU cloud providers? Include the sticker price, spot or committed discounts, egress fees, and current availability for each provider.”

22 GPU cloud providers, normalized to one H100 GPU-hour in USD from public pricing pages and comparison sources scanned through late August 2026 (report dated 2026-09-02). Effective $/GPU-hour = compute price + (egress GB generated per GPU-hour × egress $/GB); it is modeled only where a provider states an explicit egress rate or an explicit free-egress policy — a blank egress field is never treated as zero.

Lowest usable public compute price per H100 GPU-hour, ranked

zero / free egress stated explicit egress $/GB — effective price rises with traffic egress not found / contract-specific host-dependent (Vast.ai) ◇ spot/marketplace = interruptible · ● committed term required

What the ranking cannot say

Only six providers state zero or free egress — Voltage Park $0.94, TensorDock $1.50 (1-month term), Cudo Compute $1.82, Lambda Labs $1.893 (1-year reservation), GMI Cloud $2.00 and Spheron $2.91 — making them the only rows whose effective price is flat at any traffic level. voltagepark.com

AWS's $0.09/GB egress turns a $1.90 committed compute hour into $2.80 at 10 GB/hour and $10.90 at 100 GB/hour — egress alone can quintuple the bill for serving-heavy work. aiwiki.ai

Nebius charges $0.015/GiB on Standard Object Storage egress ($2.30 at 10 GB/h, $3.65 at 100 GB/h on its $2.15 rate) but its Enhanced tier egress is free — the tier choice, not the GPU rate, decides the effective price. aiwiki.ai

Two rows look inconsistent and are shown as observed rather than reconciled: Google Cloud's committed figure ($12.00) exceeds its $11.06 sticker, and Spheron's spot observation ($2.91) exceeds its $2.01 sticker — verify the linked calculators before purchase. aiwiki.ai

The cheapest rates are conditional: Voltage Park's $0.94 is a discounted/spot observation, Vast.ai's $1.49 comes from an unverified community host, and Azure's $1.40 is a preemptible spot rate — treat all of them as interruptible or inventory-dependent, not guaranteed capacity. spheron.network

Which discount to buy, by workload

Egress-heavy inference

Pick self-service zero-egress providers — Voltage Park, TensorDock, Cudo, Lambda — where the compute rate is the whole bill.

Steady utilization

Committed rates win: TensorDock $1.50/mo term, CoreWeave $1.79, Lambda $1.893/yr, AWS $1.90/yr. Commitment length is the price.

Checkpointable jobs

Spot and marketplace only — Voltage Park $0.94, Vast.ai $1.49, Azure spot $1.40 — capacity can be reclaimed at any time.

Enterprise reach

Hyperscalers ($6.88–$12.29 sticker range here) when ecosystem, controls and regions outweigh the compute and egress premium.

Every provider, every price

Provider	Representative offer	Sticker $/h	Spot $/h	Committed $/h · term	Lowest $/h	Egress	Eff. @10 GB/h	Eff. @100 GB/h	Availability	Detail	Date	Src	Source

Data: 22 canonical provider rows selected from a 98-row scan of public H100 pricing pages and comparison sources; latest source dates 2026-06-10 to 2026-08-24; report dated 2026-09-02. All prices normalized to USD per single H100 GPU-hour, though products differ (PCIe vs SXM/HGX, VM vs bare metal, frequent 8-GPU minimums). Effective price = compute + modeled egress; blank egress is shown as "not found", never zero; Vast.ai networking is host-dependent and left indeterminate. "Current availability" reflects what scanned pages stated on the listed date, not live capacity. Category/alias rows and non-shortlist providers were cut for space.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.