The National Compute Grid plans to match AI workloads with capacity pooled across providers, citing unused data-center resources and limited access for smaller teams.
A coalition of AI startups, cloud providers, researchers and investors is launching the National Compute Grid, a shared system intended to connect organizations seeking AI computing with capacity that might otherwise sit idle. Axios reported the initiative on October 7, citing interviews with participants and a paper released alongside the announcement.
The proposed grid would pool computing from AI labs, cloud providers and different chip systems, then use a shared scheduler to match workloads to available resources. Participants would be able to view information such as chip type, location, price and utilization, contribute unused capacity, or reserve larger clusters for planned training runs. The organizers say about 760 megawatts are connected or in sight, with a goal of reaching 2 gigawatts by 2030.
The paper cited by Axios says independent, single-tenant data centers average less than 15% net computing utilization. That figure is the consortium’s claim, not an independently established measure of all AI infrastructure. Anjney Midha, who is leading the effort, described the grid as a way to coordinate capacity for commercial research and public-sector use. The project says it is opening access to teams and employees in government, education and national laboratories.
If the system works as intended, it could give smaller companies and research groups another route to compute, rather than relying solely on large providers or long-term capacity contracts. 1X’s head of AI, Sam Sinha, told Axios that access can be difficult for companies unable to match the commitments larger labs can make. But the launch does not establish how much capacity will be available in practice, how pricing will work across providers, or whether workloads can move smoothly among different chip platforms. The grid’s stated capacity goal is a plan, not a report of capacity already operating through the system.