Firm electricity becomes an explicitly allocated input, and computation relocates to where it is
Regulators stop treating very large loads as ordinary customers and create a distinct class with its own rules: curtailment-enabled interconnection,…
Claude · 2032–2042 · likely
Prior state
Load growth from computation has collided with interconnection queues, transformer and turbine lead times, local siting opposition and rising retail bills. Utilities and hyperscale buyers have signed long contracts whose delivery obligations fall in the early 2030s.
Material change
Regulators stop treating very large loads as ordinary customers and create a distinct class with its own rules: curtailment-enabled interconnection, bring-your-own-generation requirements, ratepayer indemnification, and cost allocation that insulates households. Because those rules differ sharply across jurisdictions, siting decisions follow surplus firm power and permissive permitting rather than proximity to users, dispersing large-scale computation toward the Gulf, India, Brazil, the Nordics and the interior of North America.
Why now
Contracts signed in the late 2020s come due; large-load tariff dockets are filed and decided on regulatory timetables of two to four years; and the new firm supply ordered in the mid-2020s — turbines, reactor uprates, geothermal — arrives on lead times that centre on this window. Electricity prices become an electoral issue in several democracies at the same moment, which forces the allocation question into statute rather than leaving it to bilateral deals.
Mechanism and resistance
Ratepayer advocates and state legislators resist socialising the cost of private load. Water use for cooling becomes a siting constraint in arid regions. National-security reviews restrict where sensitive workloads may run, which cuts against pure energy economics. Efficiency gains partially offset demand, and a portion of announced capacity is never built.
Consequences
Energy-exporting states become computation-exporting states, converting stranded gas, solar and geothermal into a tradeable service — a partial and unequal answer to the petrostate problem described below. Sovereignty politics attaches to compute location: jurisdictions demand local processing of their own data even where power is dearer. Domestically, the settlement is regressive or progressive depending entirely on the cost-allocation rule, and this becomes one of the decade's recurring distributive fights.
End state
A formalised two-tier interconnection regime in most large electricity markets and a materially more dispersed map of large-scale computation than existed in 2030.
Observable test
Existence of regulatory tariff classes for loads above a defined megawatt threshold in named jurisdictions, and the share of new large-scale computing capacity sited outside the United States, the European Union and China.
Disconfirming sign
Efficiency improvements collapse load growth so that interconnection ceases to bind and large-load tariff classes are withdrawn.
Themes
AI & compute, Energy & resources, Infrastructure & transport