The cheapest H100s are spot bets: verified provider quotes start near $1.33, while marketplace floors dip to $0.34

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.”

15 GPU cloud providers scanned across two live-web passes (1,673 broad results and 2,921 provider-domain pages), current through 2026-09-01. All prices are USD per physical H100 GPU-hour; observed discounted floors span $0.34–$10.00, but the lowest are volatile spot or scope-ambiguous observations, not durable quotes.

stickerspot / marketplace floor (interruptible)committed floor▲ flagged: conflicts with provider-published rows

Effective cost at 10 GB of public-internet egress per GPU-hour

effective $/h = lowest observed compute rate + 10 GB × first-paid-tier egress $/GB

Only providers with an exact egress rate found are shown; explicit zero egress counts as zero. Substitute your own GB/hour mentally — every extra GB/hour adds the per-GB rate. Providers without a resolved egress rate stay unresolved, not zero.

What the visual cannot say alone

Vast AI's $0.34 spot floor is 86% under its own $2.40 sticker — but it is variable marketplace stock, never guaranteed capacity. nodepedia.com

AWS's $0.74 extracted floor is unverified: AWS's own Capacity Blocks pages price H100 at $4.72–$5.191 per GPU-hour and require a capacity reservation. aws.amazon.com

Google's $0.439 candidate came from a page whose product scope may not be H100; official accelerator pricing puts H100-family spot-equivalents at roughly $4.30–$7.01 per GPU-hour. cloud.google.com

Hyperstack's $0.53 broad-search figure conflicts with its own site: $1.33 committed for H100 PCIe and $2.04–$2.72 committed for H100 SXM — the defensible rates. hyperstack.cloud

CoreWeave's provider page supports about $6.155 on-demand and $2.439 spot per H100-hour in Europe; the extracted $0.77 may reflect unit ambiguity. coreweave.com

All 15 providers — click a header to sort

Provider	H100 offering	Sticker $/h	Spot $/h	Committed $/h	Lowest observed $/h	Egress $/GB	Effective @10GB/h	Availability	Source

Methodology

A “real” rate must separate verified provider quotes from volatile or scope-ambiguous web observations. Spot and marketplace prices are interruptible and fluctuate; committed rates need contracts of months to years (AWS Capacity Blocks are paid up front at reservation). H100 PCIe and SXM/NVLink are not performance-equivalent even though pricing is normalized per physical GPU-hour. Availability is categorical, not an SLA; blank means not publicly resolved in this scan. Egress shown is the first paid public-internet tier — actual bills depend on destination, region, and monthly allowance (e.g. AWS 100 GB/month free, Vultr 2 TB/month free, Azure 5 GB/month).

Data: two live-web scans (1,673 broad + 2,921 provider-domain results), 15 provider summary rows validated against 53 provider-domain rows, current through 2026-09-01; per-row source dates 2026-05-12 to 2026-08-28. Prices are USD per physical H100 GPU-hour, excluding storage, CPU/RAM bundle differences, taxes, support, and inter-region traffic. Regions and variants were not averaged; granular per-region rows were cut for space and used only for validation.

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.