The constraint on sovereign AI turned out to be the power bill
New York paused permits for any data centre drawing 50 megawatts or more. Australia said new ones must put back at least as much energy as they take. Sovereign compute is being repriced by electricity regulators, not by chip supply.
On 14 July, Governor Kathy Hochul signed Executive Order 62 and made New York the first state in the US to halt new hyperscale data centres. It is a one-year pause on the discretionary state environmental permits any facility drawing 50 megawatts or more needs, and it runs until the Department of Public Service delivers a generic environmental impact statement on what these buildings do to energy demand, water and air.
A day later, on the other side of the world, the Australian Prime Minister set out a condition rather than a pause: large-scale AI data centres would be required to put at least as much energy into the grid as they take out of it.
Neither government is trying to stop the build. Both are repricing it, and the direction of travel is the same. For two years the sovereign AI conversation has been a conversation about chips, on the assumption that a country with enough GPUs has enough sovereignty. These two decisions say the binding constraint is somewhere else entirely: electricity, water, and a permit signed by someone whose job is the grid rather than the technology strategy.
The practical consequence is a timeline problem, and it lands on the buyer rather than on the builder. If your data residency commitment rests on a facility that has not been built yet, your compliance date is now downstream of an environmental review in a jurisdiction you do not control and cannot accelerate. A region on a roadmap is not a region.
It is one of the plainer arguments for deploying into infrastructure that already exists. A model running on hardware a client already owns is sovereign on the day it ships, and its schedule does not depend on anybody clearing a permit queue.