A coalition of AI startups, cloud providers, researchers and investors is launching the National Compute Grid, Axios reports. The project is intended to connect computing capacity that might otherwise sit idle and make it available to companies, researchers and government teams.
The premise is that AI computing is scarce but not always busy. The consortium says independent, single-tenant data centers average less than 15% net utilization. Training runs can consume nearly all of a company's capacity, while expensive chips may be underused between those bursts.
A shared scheduler for different chips
The grid plans to pool capacity from AI labs, cloud providers and systems using different processors. Members would see available capacity, chip type, location, price and utilization, then use a shared scheduler to match workloads with suitable hardware.
Participants could contribute idle capacity, buy computing time as needed or reserve larger clusters for planned training runs. Axios reports that the consortium says it has about 760 megawatts connected or in sight, with a target of 2 gigawatts by 2030. Those figures are the group's own measures, not an independent audit of usable capacity.
A public-sector commitment
POLITICO reports that National Compute plans to donate $100 million in computing credits to the Trump administration's Genesis Mission, a government effort to use AI in scientific research. The announcement was expected at a Thursday event with White House science and technology policy director Michael Kratsios. The White House declined to comment to POLITICO.
The credits are intended to give Genesis Mission researchers access to advanced hardware that smaller labs can struggle to secure. POLITICO says National Compute plans to begin letting public-sector users with .gov or .mil addresses, along with students and faculty using .edu addresses, request access through the scheduling system starting this week.
Testing models on the grid
National Compute is also introducing an invite-only Grid Defense program. According to POLITICO, an agent called Marshall would coordinate tests defined by government agencies, keep those procedures confidential and return anonymized results to researchers.
The program does not amount to government approval of a model. Labs would decide voluntarily whether to release their systems, and the project would still need participation from both AI companies and federal agencies.
