Compute Exchange adds resale marketplace for Nvidia H100 and A100 GPUs
The GPU procurement platform is adding physical hardware sourcing as AI buyers seek lower-cost capacity beyond Nvidia’s newest chips.
By Dominic Okoye · Staff Writer
· 3 min read
Compute Exchange Inc. has launched a marketplace for used and refurbished Nvidia GPUs, with an emphasis on H100 and A100 systems. The move extends the company from reserved GPU capacity procurement into physical AI hardware sourcing, a sign that demand for production-grade accelerators is no longer confined to the newest Nvidia platforms.
The company did not disclose pricing, transaction fees, supplier count for the new marketplace, revenue, funding or valuation. Compute Exchange said buyers have already made requests ranging from hundreds of GPUs to tens of thousands, spanning AI startups, cloud providers, enterprises and other infrastructure operators.
The offering is built around matching hardware buyers with qualified suppliers. Compute Exchange said buyers submit requests through the platform, which then supports sourcing, logistics and transaction execution. The company also said it operates in a market-neutral position between buyers and sellers, a claim meant to distinguish it from more traditional procurement channels where inventory, pricing and incentives can be less clear.
The target inventory is older-generation Nvidia hardware, particularly the H100 and A100. Compute Exchange said demand in the used market is concentrated around hardware already proven in production, rather than around the newest GPU generations. That is a practical read of the AI infrastructure market: Nvidia’s Blackwell systems are reaching customers, but many workloads still run on H100 and A100 capacity.
Carmen Li, chief executive of Compute Exchange and founder of Silicon Data Inc., said in a statement that “not every workload requires the newest generation of GPUs.” She said used and refurbished hardware can offer organizations a faster and more economical way to expand AI infrastructure, and described the resale market as part of making AI infrastructure “more efficient, transparent and accessible.”
The H100, introduced in 2022, and the earlier A100 were central to the first wave of generative AI deployments and remain common in data centers. Compute Exchange is betting that buyers who cannot justify top-end systems, or cannot get them quickly enough, will treat used accelerators as a legitimate capacity option rather than a stopgap.
Compute Exchange originally focused on reserved GPU capacity, giving AI teams a procurement venue for compute across more than 100 providers. The new hardware marketplace adds physical sourcing to that capacity-discovery business, broadening the company’s role from helping customers book compute to helping them acquire infrastructure outright.
Silicon Data, founded by Li, provides real-time GPU pricing benchmarks and indexes used to value compute. Li became chief executive of Compute Exchange in October. The company was co-founded by Don Wilson, founder of trading firm DRW Holdings Inc., and Suna Said, chief executive of Woodside AI Venture Studio LLC.
Compute Exchange said the marketplace is now accepting requests from organizations looking to buy or sell used and refurbished GPU infrastructure. What remains undisclosed is how much inventory is available, how pricing will be benchmarked in live transactions, and whether enough sellers will participate to make the market meaningfully liquid.
This story draws on original reporting from SiliconANGLE.
