Amp Infrastructure wants to be the grid operator of frontier AI compute
A new outfit called Amp Infrastructure is pitching itself as a neutral coordinator for shared AI compute, borrowing its governance model from the electric grid.
Amp Infrastructure is building shared compute capacity for frontier AI teams at gigawatt scale, and its founding logic is that no single cloud provider should be the one running the room. The model it has borrowed comes from electric grid governance: an independent system operator sits above competing generators and coordinates capacity without owning the power itself. Anjney Midha describes Amp’s role in exactly those terms.
We see ourselves as what's called an independent system operator. Anjney Midha
The scale Midha projects is striking. On his estimates, frontier AI teams will need a steady-state base load pool of 1.3 gigawatts, plus roughly 6 gigawatts of spike capacity over the next four years. That is a supply problem as much as a coordination one, and Amp’s neutral-coordinator framing is partly an answer to it: teams that might resist sharing capacity with a rival cloud may be more willing to pool with a declared non-competitor.
There is a demand-side constraint worth watching, too. Midha estimates that up to 20 percent of all US data centers in 2026 are at risk of cancellation or delay because of community backlash. That is a single speaker’s estimate, not a confirmed industry figure, but if it is even directionally right it adds urgency to the question of whether new gigawatt-scale infrastructure can actually get built on the timeline frontier AI teams are counting on.