ZDI - Zevin Intelligence News

ZDI - Zevin Intelligence News

The Zevin Intelligence Journal

Chanan Zevin | Chief Editor and Head of Desks

Editorial

Morning Edition · Fri 25 Sept 2026

AI Ambition Without Power Is Incomplete Strategy

Compute plans that ignore grid interconnection and firm power will transfer risk to communities and investors.

Contextual photograph for this editorial
File photograph, not of the events described. Photo: Robert Wetzlmayr / Wikimedia Commons, CC BY-SA 4.0

The Zevin Intelligence Journal

Published

Power constraints are no longer a footnote to artificial intelligence. They are becoming a binding condition on how quickly compute can expand, where it can sit, and what societies will pay for the privilege. That is the concrete development that matters now: plans for training and inference capacity are colliding with the slower realities of generation, transmission, and grid interconnection.

Our thesis is direct. Responsible AI scale-up requires treating electricity, interconnection queues, and siting rules as first-order industrial strategy—not as afterthoughts for facilities teams. Without that shift, firms will chase chips while communities absorb outages, higher rates, or opaque special deals, and public trust in both technology and utilities will erode.

What the evidence shows Public energy research and official energy policy materials make the direction of travel clear even when exact project-level numbers vary. Electricity demand linked to digital infrastructure and data centers is rising as a planning concern for governments and system operators. The United States energy establishment likewise frames grid modernization, reliability, and generation adequacy as strategic priorities for a digital economy. These are established institutional signals, not market tips.

Inference follows. If compute demand rises faster than permitted interconnection and firm power, three outcomes become likely: delayed facilities, concentrated geographic bottlenecks, and political conflict over who pays for upgrades. Forecasts of exact megawatts will be revised; the causal chain will not. Normatively, publics are entitled to clarity about reliability trade-offs before projects are locked in.

Consequences For businesses, opaque power risk becomes balance-sheet risk—capex stranded by queue delays or contract repricing. For markets, uncertainty about energy availability can reprice infrastructure names more violently than model-release news. For institutions, fragmented permitting without transparent reliability metrics invites capture and backlash. For people, the cost of unplanned demand spikes can appear as higher bills or weaker service quality in regions asked to host facilities they did not design.

The strongest objection The strongest counterargument says markets will solve it: higher power prices will draw generation and storage, and hyperscalers already sign long-term contracts that fund new capacity. That mechanism is real and should be welcomed. It does not fully answer the timing problem. Interconnection studies, transmission build, and community consent run on calendars measured in years, while compute product cycles run in months. Price signals help; they do not invent transformers overnight.

What should happen next First, utilities and independent system operators should publish standardized interconnection transparency for large-load applicants—queue position, study status, and reliability impacts—so communities and investors see the same facts. Second, hyperscalers and industrial buyers should pair new load announcements with verifiable firm-power or flexibility plans, including demand-response participation where grids are tight. Third, policymakers should condition accelerated permitting for large loads on measurable reliability and cost-allocation standards, not slogans.

Evidence that would change our position includes sustained multi-year data showing interconnection lead times compressing while reliability metrics improve in high-load regions. Until then, celebrating model capability without power realism is incomplete strategy.

AI will keep improving. The question is whether institutions improve the physical systems that make that progress shared rather than precarious. Ambition without electrons is not leadership; it is delay wearing a press release.

Shared responsibility None of this argues against AI investment. It argues against magical thinking about infrastructure. Communities that host large loads deserve to know whether reliability standards will hold. Investors deserve to know whether announced campuses can energize on schedule. Workers deserve training pathways that match construction timelines rather than marketing calendars.

A liberal, pragmatic line supports innovation when it expands opportunity and respects institutional accountability. That means welcoming private capital that funds generation and storage while insisting that cost allocation remains visible. It also means rejecting the false choice between technological progress and public trust. Both are required for markets to function with legitimacy.