Energy Replaces Compute as the Binding Constraint on US AI Dominance
A $1.4 trillion utility infrastructure race reveals that power—not chips or talent—now determines which nation leads in artificial intelligence.
US investor-owned utilities have committed $1.4 trillion in capital spending through 2030, a 27% increase from last year’s projection, driven almost entirely by the explosive power demands of AI data centers that now account for half of all US electricity demand growth.
The unprecedented infrastructure surge, detailed in PowerLines analysis released last month, marks a fundamental shift in how the United States plans and finances its electric grid. Data center power demand is projected to grow from 35 GW in 2024 to 78 GW by 2035, with average hourly electricity consumption nearly tripling to 49 GWh, according to BloombergNEF. That growth rate surpasses residential, industrial, and transport sectors combined—and it has caught both utilities and regulators structurally unprepared.
Jason Bordoff, founding director of Columbia University’s Center on Global Energy Policy, captured the scale of disruption:
“What we’re witnessing is a complete rewiring of the American energy system in response to a technology transition. The speed at which AI demand is growing has caught both utilities and regulators off guard.”
— Jason Bordoff, founding director, Columbia University Center on Global Energy Policy
Nuclear and Dispatchable Power Emerge as Strategic Assets
The capital deployment is not evenly distributed across generation types. Nuclear, natural gas, and other dispatchable sources are receiving the bulk of investment, while traditional renewables—despite their cost advantages—prove structurally unsuited to AI workloads that require constant 24/7 power with minimal variance. Constellation Energy’s $16.4 billion acquisition of Calpine in January 2026 exemplifies the strategic pivot: the deal added natural gas and geothermal assets specifically for ‘firming’ services to balance nuclear output.
NextEra committed to restarting the Duane Arnold nuclear plant in Iowa (615 MW) by 2029 under a 25-year power purchase agreement with Google, as reported by the Motley Fool. The company now has 6 GW of capacity either operating or contracted with Google and Meta. Meta separately signed 20-year agreements with Vistra for 2,176 MW from the Davis-Besse and Perry nuclear plants in Ohio, part of a broader 6.6 GW nuclear commitment spanning Vistra, TerraPower, and Oklo by 2035, per World Nuclear News.
The pipeline of conditional offtake agreements between data center operators and small modular reactor projects grew from 25 GW at the end of 2024 to 45 GW as of this month, according to the International Energy Agency. Yet these projects face regulatory uncertainty—NRC licensing timelines remain unpredictable, and no commercial SMR has yet achieved full-scale deployment in the United States.
Regional Grid Stress and Rate Pressure
Southern Company secured 28 large-load projects totaling 11 GW under contract as of the first quarter, with data center power consumption up 42% year-over-year across its utilities, the company reported in earnings released May 1. Dominion Energy holds 47.1 GW of data center contracted capacity, including 9.8 GW in take-or-pay agreements and another 9 GW in construction letters of authorization, per Latitude Media.
But concentration creates vulnerability. The PJM Interconnection projects potential loss-of-load hours in Northern Virginia could spike from 2.4 to 430 per year by 2030 without massive capacity additions, with 15 GW of unbuilt data center demand in the PJM South region alone, according to Lawrence Berkeley National Laboratory analysis. Morgan Stanley forecasts a 49 GW shortfall in US data center power by 2028—nearly 40% of the projected global 126 GW increase.
| Utility | Contracted Capacity | Key Customers |
|---|---|---|
| Dominion Energy | 47.1 GW | Multiple hyperscalers |
| Southern Company | 11 GW | Google, Meta, Microsoft |
| NextEra Energy | 6 GW | Google, Meta |
| Entergy | 5-10 GW | Amazon, Alphabet, Meta |
Rate increases are already materialising. Electric and gas utilities requested more than $30 billion in rate hikes in 2025, affecting 81 million Americans, PowerLines reported. A Carnegie Mellon study estimates Data Centers and crypto mining could increase average US electricity bills by 8% by 2030, potentially exceeding 25% in Virginia. Goldman Sachs warns the data center-driven demand surge will boost core inflation by 0.1% in both 2026 and 2027, with the greatest impact concentrated in PJM region states.
Local opposition is mounting. At least 16 data centers worth a combined $64 billion faced blockage or delay in 2025, and Maine approved a statewide data center moratorium, according to Fortune.
Energy as the New Geopolitical Constraint
The infrastructure bottleneck is reshaping the global AI competition in ways that favor China and disadvantage Europe. China’s total primary electricity consumption is roughly double that of the entire United States, offering a vastly larger energy base for AI infrastructure expansion, according to Americans for Responsible Innovation. The country also controls 99% of heavy rare earth element processing, critical for wind turbines, cooling systems, and high-performance computing.
While the US holds structural advantages in semiconductor design, cloud infrastructure, and AI research talent, energy is emerging as the binding constraint that could determine which nation achieves AI dominance by 2030. China’s centralized infrastructure planning allows faster deployment of generation capacity, while European hyperscalers face both higher energy costs and more severe grid bottlenecks than US competitors.
A Federal Reserve analysis concluded the US holds an advantage in electricity infrastructure quality but that China has a comparative advantage in energy availability for AI expansion. France generates roughly 75% of its electricity from nuclear—a strategic asset US policymakers increasingly cite as critical to AI competitiveness and national security.
Google CEO Sundar Pichai acknowledged the constraint directly:
“Scaling compute capacity while managing the power, land, and supply chain constraints necessary to meet demand for AI services remains a persistent concern.”
— Sundar Pichai, CEO, Google
What to Watch
Nuclear regulatory timelines will determine whether the SMR pipeline translates into operational capacity or remains speculative. The NRC has yet to approve a commercial SMR design for full deployment, and licensing delays could push projects beyond 2030—leaving utilities reliant on natural gas to fill the gap.
PJM capacity auction results in the coming months will reveal whether pricing signals are sufficient to attract the investment needed to prevent grid stress in Northern Virginia and other concentration zones. If capacity prices do not rise enough to incentivise buildout, the region faces either severe reliability degradation or mandatory load curtailment for data centers.
State-level regulatory decisions on cost allocation will shape whether data center operators bear the full infrastructure cost or whether it is socialised across all ratepayers. Virginia, Georgia, and Texas—the three largest data center states—are each considering different models, and those choices will determine both competitiveness and political sustainability.
Finally, China’s energy infrastructure deployment pace bears close monitoring. If Chinese utilities can bring large-scale generation online faster than US counterparts—a plausible scenario given centralized planning authority—the energy advantage could offset US leads in AI research and semiconductor design, fundamentally altering the competitive balance by decade’s end.