Compute Exchange
Marketplace for buying and selling GPU compute capacity and hardware.
Compute Exchange operates a marketplace that connects buyers and sellers of AI computing power. It facilitates the procurement of reserved cloud capacity and trades in used or refurbished graphics processing units. The platform aims to bring transparency and standardised pricing to the infrastructure market for artificial intelligence workloads.
- Launches secondary market for used Nvidia GPUs in July 2026.
- Connects buyers with suppliers of H100 and A100 hardware.
- Uses SiliconMark for independent verification of GPU condition.
- Headquartered in New York with a global seller network.
What it is
Compute Exchange is a digital marketplace designed to simplify how organisations acquire artificial intelligence computing resources. Based in New York, it connects those who need graphics processing unit power with suppliers who have available capacity or hardware. The platform handles both reserved cloud compute and physical equipment transactions. It aims to provide clear pricing and unbiased options for buyers ranging from startups to large enterprises. By acting as an intermediary, it seeks to make the procurement process more efficient and transparent than traditional direct sales channels.
How it came about
The company was founded by Don Wilson and Suna Dehghanpisheh. Wilson previously established DRW and has experience in derivative trading and venture capital. Dehghanpisheh founded Woodside AI and has invested in several technology startups. Compute Exchange launched initially to facilitate the buying of reserved GPU capacity. In July 2026, it expanded its services by introducing a secondary market for used and refurbished hardware. This move responded to growing demand for cost-effective infrastructure solutions among AI developers and cloud providers.
The compute
The platform supports transactions for Nvidia H100 and A100 graphics processing units. These chips remain popular for production workloads despite the arrival of newer generations. Compute Exchange does not own the hardware but verifies its condition through SiliconMark, an independent verification service. This allows buyers to assess quality before purchasing. The marketplace covers both older generation accelerators and newer systems, focusing on proven performance rather than just the latest specifications. It enables trading of physical units as well as cloud-based compute time.
Where the GPUs are
Compute Exchange operates as a virtual marketplace headquartered in New York. It does not list specific data centre locations because it aggregates supply from various providers. The platform connects buyers with a global network of sellers including cloud providers and infrastructure operators. This distributed model allows users to source hardware from multiple regions without dealing with each supplier individually. The focus is on the transaction layer rather than physical hosting facilities.
Customers and deals
Buyers on the platform include artificial intelligence startups, enterprise organisations, and cloud service providers. These groups seek to expand their infrastructure capacity without waiting for new hardware releases. Sellers range from companies upgrading their fleets to operators with surplus inventory. The marketplace serves those looking for hundreds or tens of thousands of units. It caters to users who prioritise cost efficiency and immediate availability over having the newest chip architecture.
Money and ownership
The company operates as a private startup. It generates revenue by facilitating transactions between buyers and sellers of compute resources. The founders have backgrounds in financial markets and venture capital, which influences the platform's approach to pricing and contracts. No specific funding amounts or valuation figures are publicly disclosed in the available records. The business model relies on making the market more liquid and transparent for all participants.
Its place in AI
Compute Exchange positions itself as the first open exchange for compute resources. It aims to treat computing power as a tradable commodity similar to financial assets. By providing real-time price discovery and standardised contracts, it seeks to reduce friction in the AI infrastructure sector. The platform also highlights carbon-efficient choices to support sustainability goals. It plays a role in democratising access to high-performance computing for innovators who might otherwise face barriers to entry.
Questions people ask about Compute Exchange
What types of hardware can be bought on Compute Exchange?
The marketplace offers reserved cloud compute capacity and physical graphics processing units. It specifically lists used and refurbished Nvidia H100 and A100 chips. Buyers can request specific hardware configurations and receive matches from qualified suppliers.
How does the platform ensure the quality of used GPUs?
Compute Exchange uses SiliconMark to provide independent verification of GPU condition. This service allows customers to assess the quality and status of the hardware before finalising a transaction. It helps build confidence in buying secondary market infrastructure.
Who are the typical users of this marketplace?
Users include AI startups, enterprise companies, cloud providers, and infrastructure operators. These groups use the platform to find cost-effective ways to expand their computing capacity. They often seek proven hardware rather than waiting for the latest generations.
Good reading and sources
There is no English Wikipedia article on Compute Exchange yet, so the facts on this page come from the sources listed above: compute.exchange/about, einpresswire.com/article/927423264/compute-exchange-launches-secondary. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.