Loading benchmark data
Historical cloud rental benchmarks for H100, H200, B200 and A100 80GB across Amazon Web Services, Microsoft Azure and Google Cloud. Built only from prices the providers publish themselves.
Provider-balanced cross-region medians, Aug 2026. One GPU, one hour, Linux, on-demand.
List prices move in steps and then hold, sometimes for a year. The chart shows exactly that: no smoothing, no invented daily ticks, and a visible break wherever a provider does not publish.
Feb 2024: Google Cloud enters the benchmark. A step here is a change in what is being measured, not a price move.
Aug 2026: Azure enters the benchmark. A step here is a change in what is being measured, not a price move.
Azure publishes no price history, so its series begins in Aug 2026, when GPUQuant recorded its first snapshot. It cannot be backfilled.
Prices are Linux on-demand list prices for the H100 80GB SXM5, eight-GPU node, normalized to one GPU-hour. How the benchmark works
Renting an NVIDIA GPU has standardized to the point where the hour is a commodity unit. CME Group and Silicon Data plan to launch H100 and B200 rental-index futures on October 5, 2026, pending regulatory review. Each contract represents roughly one GPU-month of rental capacity, the same 730 hours GPUQuant quotes.
GPUQuant is not that index and does not trade it. What GPUQuant publishes is the cash-market context underneath it: what the three largest cloud providers publicly charge to rent the same GPU, month by month, region by region.
GPUQuant figures are hyperscaler list-price benchmarks. They are not Silicon Data settlement prices and not traded CME futures quotes.
CME Group and Silicon Data plan to launch H100 and B200 rental-index futures on October 5, 2026. The launch remains subject to regulatory review and the contracts are not trading today.
Planned CME-linked product: Silicon Data H100 Rental Index Futures
Planned CME-linked product: Silicon Data B200 Rental Index Futures
Every figure traces back to a price the provider published. Third-party pricing services are deliberately out of scope.
AWS publishes a dated version of the full price list roughly once a month, archived back to December 2015. Every GPUQuant point is one of those official versions.
Provider documentation ↗Azure publishes current prices only, with no historical archive of any kind. The Azure series begins when GPUQuant took its first snapshot and cannot be backfilled.
Provider documentation ↗Google Cloud answers historical queries one calendar month at a time, back to January 2017. Every GPUQuant point is the price Google reported as effective in that month.
Provider documentation ↗A concise monthly update when official GPU rental prices change. Nothing when they do not, which is most months.