Silicon Data CEO Carmen Li says demand for older AI chips remains high. Her reasoning is simple: chip shortages are keeping prices for Nvidia’s H100 elevated, so buyers are still lining up for previous-generation silicon.
Why older GPUs refuse to depreciate
Silicon Data’s figures put H100 residual values at approximately 61% of their purchase price as of mid-October 2026, after more than four years on the market.
The older A100 tells a similar story. After six years, it still holds about 22.5% of its original value.
Li noted that on-demand H100 rentals sit in the range of $2 and up per hour.
Newer B200 GPUs, by comparison, rent for about $5 per hour.
Silicon Data launched its H100 Rental Index on May 27, 2025, with the average starting at approximately $2.37 per hour.
From chip making to chip economics
Li started Silicon Data in April 2024. The company focuses on pricing indices and market intelligence for AI compute, tracking GPU rental rates and how older Nvidia chips hold their value.
Investors appear interested in that question. Silicon Data secured $30.5 million in Series A funding in August 2026, in a round led by Valor Atreides AI Fund.
What this means for the AI compute market
The company is laying the groundwork for cash-settled GPU futures and options with CME Group, intended to launch in fall 2026, intended to give market participants a way to manage the significant price volatility in H100 rentals.
Li has outlined a potential addressable derivatives market for compute that could range from $6 to $30 trillion annually. That is a wide band, and it describes a potential market rather than an existing one, so it is best read as Li’s view of the opportunity, not a forecast.
The key variable to watch is supply. Li’s argument links older-chip demand directly to shortages, so any easing in GPU availability could change the math on both rental rates and resale values.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.