“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.”
This compares 15 GPU cloud providers with verifiable first-party H100 price pages as of September 8, 2026 — a broad current market snapshot, not every H100 seller globally, with each provider row backed by its own independent pricing host. All rates are USD per H100 GPU-hour; the ranking uses each provider's lowest currently published compute rate among sticker, spot, and committed options. Egress is counted only where the provider publishes a fee (a zero means explicitly free egress); a blank egress cell means total effective cost is not calculable from published data, not that it is free.
| Provider | Offer / variant | Sticker $/hr | Spot $/hr | Committed $/hr | Term | Best $/hr | $/GPU-mo (730h) | Egress $/GB | Eff. +1 GB/hr | Eff. +10 GB/hr | Availability | Conditions | Source |
|---|
Method: 15 provider rows compiled 2026-09-08 from 15 independent first-party pricing hosts (at most one final row per pricing URL; multiple official pages scanned for most providers). Numbers are USD per NVIDIA H100 GPU-hour as published; "best" is the lowest of sticker, spot, or committed. Effective columns add published internet egress at 1 or 10 GB per GPU-hour; free monthly allowances would be amortized over 730 hours. Blank cells are unresolved, not zero. Taxes, storage, bundled CPU/RAM, idle time, support, and regional variation are not normalized; H100 variants (PCIe, SXM5, HGX) and interruptibility differ. Some availability text reflects offer listings, not live inventory. Egress source links were cut for space; each row links its pricing source.