Sovereign AI and the Grid

Glossary

Anchor offtake

A long-term contract, for the electricity (a power purchase agreement) and often for the compute output, that commits a buyer to pay whether or not full capacity is used. It gives utilities and lenders the revenue certainty needed to justify grid investment and financing.

Behind-the-meter generation

On-site electricity generation, typically gas turbines, built at a campus to serve its load directly instead of drawing from the public grid. It bypasses the load interconnection queue but shifts the constraint to scarce turbines and fuel logistics.

Binding constraint

The single input whose scarcity sets the ceiling for an entire system, so that adding capacity elsewhere yields no gain until it is relieved. For sovereign AI the binding constraint is firm electric power delivered to a specific site on a specific date.

Cost allocation

The regulatory decision on whether the grid upgrades a large datacenter load requires are paid by that datacenter or socialized across all ratepayers. Socialized cost is cheap for the developer but burdens households; isolated cost protects ratepayers but can undermine project economics.

Interconnection queue

The grid-operator process in which new load or generation is studied for its impact on the network and, if accepted, assigned any upgrades it triggers before it can energize. On congested grids the queue stretches years and functions as a filter on which projects are real.

Large power transformer

A bespoke, oil-filled machine that steps electricity between transmission-grid voltages and the levels a substation and campus use. Built to order by a few factories with full order books, its multi-year lead time frequently governs when a campus can be energized.

Sovereign AI

A national capability to train and serve advanced AI models on infrastructure a country controls, sited in its own jurisdiction and running on power it can guarantee. Operationally it is control of the physical compute stack, not domestic chip fabrication or a from-scratch model.

Stranded-asset risk

The risk that infrastructure built to serve a large load becomes unused if that load ramps slowly or cancels, leaving the utility with costs it cannot recover. It drives utilities to demand take-or-pay terms and regulators to scrutinize how such costs are allocated.