Allen Says AI Buildout Already Underway Will Continue

Allen said AI data-center projects already underway are unlikely to be disrupted by political scrutiny and could add tens of gigawatts of compute capacity over the next 18 months. He expects astonishing growth in AI compute capacity in 2027 and 2028, while future projects may face requirements tied to tax breaks, union labor and protections against higher power bills.

Analysis

The investable distinction is between committed construction and the next wave of projects: political scrutiny is more likely to change the cost, siting and returns of future capacity than to stop work already financed or permitted. That shifts near-term exposure toward grid interconnection, transformers, switchgear, power generation and construction capacity—not an indiscriminate bet on every AI-related asset. Utilities may benefit from load-driven investment, but only where regulators let them recover costs without triggering affordability backlash; higher power bills can turn a demand tailwind into delayed approvals or tougher rate treatment.

Over 1–3 months, watch for utility load forecasts, interconnection queues, project cancellations and hyperscaler capex guidance. The “tens of gigawatts” and 2027–28 outlook are Allen’s projections, not verified delivery commitments. Over 6–18 months, execution bottlenecks and power availability may matter more than political headlines. A further risk is that capacity arrives faster than monetization: if AI revenue fails to support the buildout, customers could defer later phases even while near-term construction continues. The consensus may underweight local ratepayer and permitting friction, but near-term disruption to projects already underway may be overstated.

Key Decisions for Investors

• Prefer a selective, staged long in grid equipment, interconnection and power-infrastructure exposure over a broad AI-capex trade; add only as order books and utility investment plans confirm demand. The thesis weakens if hyperscalers cut capex or equipment backlogs roll over.
• Treat regulated utilities as a stock-specific trade, not a sector-wide beneficiary: favor those with identifiable data-center load and credible cost-recovery frameworks; avoid assuming all incremental investment earns acceptable returns.
• Do not short existing data-center construction solely on political scrutiny. Reassess if local rules materially delay permits, impose costly power-bill protections, or if developers disclose cancellations and schedule slippage.
• Set alerts for utility load-forecast revisions, interconnection timelines, power-price moves, hyperscaler capex guidance and proposed tax-break or labor requirements. These are better confirmation points than political commentary alone.