The battle for megawatts: why restricting data centers is a strategic mistake

The commentary argues governments should expand power generation and grid capacity rather than restrict data centers, warning that permitting pauses may push AI investment abroad and increase reliance on foreign infrastructure. U.S. data centers could face an estimated 49 GW power shortfall by 2028, while transformer lead times exceed two years versus roughly 50 weeks before the pandemic. The author cites nearly 12 GW of potential data center demand in New York’s interconnection queue as of May 2026 and recommends coordinated investment in grids, generation, storage, and cooling.

Analysis

The investable signal is not simply “more data centers”; it is a shift in bargaining power toward scarce, deliverable electricity and equipment. Hyperscalers may have GPUs and capital but cannot monetize either until sites, grid connections, transformers, cooling, and firm power arrive together. That can defer cloud capacity and revenue conversion even as AI demand remains strong. NVDA is therefore exposed to deployment timing, not necessarily end-demand destruction; watch customer shipment acceptance and capex-to-capacity conversion rather than GPU order headlines.

GEV and Siemens Energy (ENR) have favorable demand visibility, but equipment orders and slot reservations are not the same as near-term recognized revenue. Long factory lead times constrain volume and create execution, cancellation, and valuation risk. Transformer makers, grid contractors, storage, and liquid-cooling providers are potential secondary beneficiaries, but the article supplies no company-level evidence to underwrite specific names.

MSFT’s power procurement can protect strategic capacity, but dedicated generation and long-term contracts may raise capital intensity and lock in supply before project economics are certain. GOOG’s nuclear strategy is a longer-dated option, not near-term capacity relief. Permitting pauses can redirect investment across regions, yet local connection and equipment constraints may matter more than national policy headlines.

Contrarian point: aggregate electricity supply is not the key metric; location, transmission, and time-to-connect determine who can deploy. Onsite gas is not a quick workaround when turbine slots and permits are scarce. Over 1–3 months, track utility connection timelines, equipment order conversion, and hyperscaler capacity disclosures. Over 6–18 months, the thesis strengthens if grid investment and firm-power additions accelerate; it weakens if AI infrastructure spending is cut or planned projects are delayed or canceled.

Key Decisions for Investors

• Consider a staged long in GEV and ENR on weakness, rather than chasing order headlines. The catalyst is conversion of backlog/slots into deliveries and revenue; cap position size for long lead times and execution risk. Falsify if order cancellations rise or delivery schedules slip without improved pricing or margins.
• Avoid treating NVDA demand as immune to power constraints: monitor shipment acceptance, customer deployment schedules, and data-center capacity disclosures. A delay thesis is weakened if customers demonstrate rapid installation and monetization despite local grid bottlenecks.
• For MSFT, treat secured power as a strategic capacity hedge, not automatically an earnings positive. Watch Azure growth against infrastructure capex and power commitments; deteriorating returns on incremental capacity would undermine the benefit. GOOG’s nuclear initiatives are a longer-horizon watch item, not a near-term trade catalyst.
• Track regional interconnection timelines, transformer delivery dates, and permitting outcomes as the key 1–3 month indicators. If connection queues clear materially or hyperscalers reduce planned builds, reassess the scarcity premium in power-equipment and infrastructure exposures.