In Summary
Artificial intelligence is about to impose the largest sustained demand shock on US electricity infrastructure in decades. By 2030, data-center power consumption is expected to nearly double, lifting the sector’s share of total US electricity demand from roughly 5% to around 9%. Although planned generation additions look sufficient on paper, data centers may absorb nearly half of projected new capacity, leaving thin margins if electric-vehicle adoption or industrial electrification accelerate faster than expected. Yet demand itself is evolving faster than forecasts can capture: generative AI reached 53% population-level adoption in just three years, agentic systems consume far more energy per interaction than conventional workloads and a 280-fold decline in inference costs since 2022 is driving a rebound effect that efficiency gains alone cannot offset.
The critical bottleneck, however, is not generation capacity but the grid itself. Data centers can be built in under two years, but grid connections can take up to seven years in congested markets such as Northern Virginia. Nationwide interconnection requests now total 1.84 terawatts, exceeding total installed US generating capacity. Supply-chain shortages have pushed lead times for critical grid equipment to several years, while projected demand would require building roughly 8,000 km of high-voltage transmission lines annually – around ten times the current pace. The strain is already showing: In early 2026, the Department of Energy invoked emergency powers to shift data centers onto backup generation during peak demand periods. Rising public opposition and legislative scrutiny add a further layer of uncertainty that conventional supply forecasts have yet to fully capture. The strain is already visible in project pipelines, with half of the 12 GW of US data-center capacity planned for 2026 being delayed or cancelled.
Aggregate electricity prices have not yet fully reflected these pressures, but a growing "data-center premium" is emerging. States hosting the highest concentration of data-center activity have so far seen price inflation below the national average, reflecting favorable grid conditions, economies of scale and the lagged structure of utility rate-setting. Beneath the surface, however, the impact has become increasingly visible since 2023. US residential customers are already paying USD1.4bn more per year on their electricity bills as a direct result of data-center demand, with just five utilities serving 4.4mn households in Northern Virginia, the Pacific Northwest and Arizona accounting for more than 40% of that total. The sales-weighted average price effect across all utilities sits at just 0.6%, but for the most exposed utilities roughly 7.8pps of a 24.5% cumulative price increase between 2020 and 2024 are directly attributable to data-center demand, adding 0.19pp to headline inflation over four years through the direct electricity channel alone. These markets historically benefited from below-average electricity costs, a gap that has already narrowed from 5% to 3.7% since 2020 and is set to close further. Data-center investment grew 32% in 2025 and is set to rise a further 75% in 2026 alone, pointing to an additional electricity price effect of close to 14pps for the most exposed utilities over 2025–2026, nearly doubling the cumulative four-year effect in just two years. Meanwhile investor-owned utilities filed USD18bn in rate-increase requests in 2025, the highest since the mid-1980s, with costs largely falling on existing rate-payers rather than the facilities driving them.
Preparing energy infrastructure for AI and addressing community concerns is as urgent as building the AI infrastructure itself. The immediate priority is interconnection reform: binding timelines, penalties for speculative queue filings and priority treatment for shovel-ready projects with firm power commitments would help relieve the most acute bottlenecks. Cost allocation is equally important. Unless data centers bear a proportionate share of the infrastructure costs they create, public opposition and permitting delays will intensify further. Mandatory energy-use disclosure, incentives to redirect investment away from saturated regions, stronger efficiency standards, demand-flexibility mechanisms and stricter additionality requirements for power-purchase agreements would fill the most critical gaps in a policy framework that remains largely inadequate to the scale of the challenge – and lay the foundation for AI ambitions that the grid can actually support.