Rival GPU price trackers put the cost of the same H100 chip roughly eight times apart, based on a July 29 table from Gpuinstances and a Silicon Data index published April 27.

The eightfold gap compares Gpuinstances' July 29 $0.94 marketplace spot listing with the $7.40 lower bound of Silicon Data's April 27 hyperscaler on-demand benchmark, a measure covering a different provider universe, different contract terms and a March 1 to April 20 window. CME's planned contracts would settle against Silicon Data's indices, giving the benchmark a direct role in the futures market. Carmen Li, Silicon Data's founder, described the pricing problem in CME Group's May 12 release. "Compute markets today are still highly fragmented, with pricing that can vary dramatically across providers, regions and contract structures," she said.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

A July 29 Show HN submission pointed to Gpuinstances, whose captured table contained 343 rows from 18 providers and covered 76 GPU models. Among its H100 offers were Prime Intellect spot at $0.94 per hour and GCP preemptible at $0.9799. Updated the same day, GetDeploying's H100 page said it covered 49 providers and placed the range at $0.35 to $14.90 per GPU-hour. It reported an average of $3.62 and a median on-demand price of $3.25. Those figures did not identify a single H100 market price.

Silicon Data's earlier measurement showed little movement within its hyperscaler series. Li's April 27 report put the H100 on-demand band between $7.40 and $7.52 from March 1 through April 20. The coefficient of variation was 0.47%, meaning the hyperscaler price barely moved across the window. Over 50 trading days, the index was unchanged on 39 of 50 daily transitions and posted a net decline of 0.27%. A threefold hyperscaler premium over neocloud rates persisted through the measurement period.

Silicon Data says its benchmark standardizes observations for differences in machines, rental terms, platform performance and geography, producing an index instead of a table of raw offers. Its product page showed the SDH100RT neocloud index at $2.53 per GPU-hour during capture. The company says its data cover 95% of tracked neoclouds and all major hyperscalers.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

GetDeploying's on-demand series moved upward during its own comparison period. It said the median H100 price increased about 8% from July 2025, rising from $3.00 to $3.25 per hour. Silicon Data's hyperscaler tier changed by two cents over its 50-day window.

CME Group announced its Silicon Data partnership on May 12 and said the first compute futures for GPU markets would launch later in 2026. Under the announcement, the contracts would settle against Silicon Data's indices. The release called them "the world's first daily GPU benchmarks for on-demand rental rates." CME's May 12 release said the launch was subject to regulatory review.

Frequently Asked Questions

Why do GPU price trackers report such different H100 prices?

They measure different things. The $0.94 figure is a marketplace spot listing, while Silicon Data's $7.40 lower bound benchmarks hyperscaler on-demand contracts across a different provider universe, different contract terms and a March 1 to April 20 window.

What is CME Group planning to launch?

CME Group announced on May 12 a partnership with Silicon Data to launch the first compute futures for GPU markets. The contracts would settle against Silicon Data's indices, with launch later in 2026 subject to regulatory review.

Are H100 rental prices falling?

Not according to these measurements. GetDeploying said its median H100 on-demand price rose about 8% from July 2025, from $3.00 to $3.25 per hour, and Silicon Data's hyperscaler tier changed by two cents over its 50-day window.

How does Silicon Data build its index?

Silicon Data says its benchmark standardizes observations for differences in machines, rental terms, platform performance and geography, producing an index rather than a table of raw offers. The company says its data cover 95% of tracked neoclouds and all major hyperscalers.

What is the gap between hyperscaler and neocloud GPU pricing?

A threefold hyperscaler premium over neocloud rates persisted through Silicon Data's measurement period. Its SDH100RT neocloud index showed $2.53 per GPU-hour at capture.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Meta Is Developing AI Cloud Plans as Shares Jump 9.3%
Meta Platforms is developing plans for a cloud infrastructure business to sell outside customers access to AI computing power and hosted models, Bloomberg reported Wednesday. Meta shares jumped 9.3% t
The Implicator
Meta Weighs Cloud Business as Capex Guide Rises to $125 Billion-$145 Billion
Meta CEO Mark Zuckerberg told shareholders Wednesday that the company could enter cloud computing if its AI data-center buildout leaves it with excess capacity. The option would turn spare compute cap
The Implicator
Zuckerberg Offers Wall Street a Cloud Answer for AI Spending
San Francisco | Thursday, May 28, 2026 Meta is no longer only buying compute for its own feeds, agents and ads machine. Zuckerberg told shareholders a cloud business is "definitely on the table" if
The Implicator
ai-news

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai