AMP PBC

AMP PBC is a public benefit corporation building an independent AI compute grid. Its stated purpose is to “maximize the world's frontier output.” Anjney Midha, introduced as its founder and CEO in a June 18, 2026 interview, described AMP’s approach as pooling compute across providers and research teams to improve utilization.[2][1]

The AMP grid is designed as a horizontal, multi-cloud, multi-silicon pooling layer. Midha compared its role to an independent system operator in an electricity grid: coordinating supply and demand so participants can share resources rather than each maintaining capacity for their own peak needs. In his formulation, neoclouds supply the grid, while research labs provide demand. The goal is to “make FLOPs flow like megawatts,” although stranded compute and limited fungibility make that difficult.[1]

AMP began at the scheduling layer and also works at the economic layer. Midha credited engineering leaders Seb and Mihai with building related scheduling systems at Google. The grid design gives participating labs guaranteed base-load capacity while allowing them to increase or decrease compute use on shorter timelines. In June 2026, he described the working implementation as involving labs, universities and trusted parties; an open protocol connecting outside supply and demand remained an ambition. His proposed two-way connection with SF Compute was a hoped-for integration.[1]

The base-load and flexible-spike design had earlier precedents. Midha linked it to Oxygen, a program he said he devised at a16z, and to Google scheduling implementations built by Mihai and Seb. To explain the Google precedent, he described interruptible jobs and a credit-based bidding mechanism that dynamically prioritized competing workloads.[1]

By June 2026, Midha described AMP as a holdings company with an infrastructure business and a capital business called Foundry, which incubates frontier AI labs or invests in them. He cited Periodic Labs as an example and said their fund had invested “a few hundred million dollars” in Anthropic earlier that year, an investment in Anthropic rather than a financing round raised by AMP.[1]

Midha described AMP’s compute procurement as favoring trusted partners with long operating records, including established data-center providers outside the AI-provider category. He connected reliable infrastructure to community support and linked AMP’s public-benefit incorporation to two priorities: net-positive data centers and empowering patients through more precise end-of-life prediction. The healthcare effort was an incubation he was working on in June 2026.[1]

AMP’s gigawatt figures in the June interview described demand and ambitions, rather than fully secured operating capacity. Midha said the desired steady state was a 1.2 GW base-load pool and estimated a need for roughly 6 GW of spike capacity over the next four years. He explicitly clarified, “we haven’t secured all of it,” and said AMP had not publicly confirmed how much capacity it had for that year.[1]