The AI power crunch is a storage problem
Hyperscale campuses are hitting grid limits well before they hit compute limits. Behind-the-meter storage is the fastest way to unlock stranded capacity.
Every hyperscaler is now grid-constrained somewhere in its build plan. Interconnection queues in the US, UK and Ireland stretch to the mid-2030s. Even where connections exist, capacity headroom is thin.
Behind-the-meter battery storage changes the physics. A campus can draw within its contracted maximum while storage carries peaks, shapes load and provides ride-through during transients. The grid connection stops being the constraint.
This is not theoretical. Operators are already deploying multi-hundred-MWh systems on-campus specifically to unblock AI training and inference build-outs that would otherwise wait years for network reinforcement.
The same 5 MWh AC block that ships to utility projects works behind the meter here. What changes is the dispatch logic, not the hardware.