Electricity prices surge 800% as AI-driven demand strains the US power grid

U.S. electricity demand is continuing to peak as artificial intelligence emerges as a new growth driver, increasing pressure on the power system. The impact is not uniform: demand is rising both in total and sharply within specific infrastructure clusters, particularly in data center–dense regions where electricity consumption has outpaced available supply.

PJM capacity prices surge as AI-linked demand strains supply

In summer 2025, capacity prices in the PJM market—where most U.S. data centers are located—rose by more than 800% year over year, signaling a tightening of supply relative to demand. As demand spikes, the system increasingly depends on readily dispatchable generation to maintain reliability.

More oil, gas and coal plants kept online

Reuters reports that in the PJM region, about 60% of plants running on oil, gas and coal that were scheduled to retire in 2025 have had those retirements postponed or canceled. Many of the affected facilities are peaker plants, which are designed to run only for short periods during peak hours.

While peaker plants are high-cost and typically more polluting, they serve as a last line of defense against wide-scale blackouts. Extending their operating lives reflects the broader strain on the grid rather than a quick technical fix.

Emissions and capacity role of peaker plants

Reuters, citing federal studies, says peaker plants may emit sulfur dioxide at about 1.6 times higher per unit of electricity than conventional plants. It also notes that roughly 1,000 peaker plants in the United States can provide up to 19% of capacity while contributing only about 3% of output.

Risk shifts from shortages to potential excess capacity

AI-driven demand growth could eventually create a reversal—from shortages to excess capacity. The Financial Times reports that in 2026, energy investors warned that building electricity infrastructure too quickly for AI could result in excess capacity if actual demand falls short of expectations or if technological performance improves faster than forecast.

Data center demand outlook raises investment needs

Demand from data centers is forecast to grow at an accelerating pace, requiring utilities to expand capacity at a faster rate than in prior cycles. Estimates cited by the Financial Times suggest electricity demand for U.S. data centers could rise from 34.7 GW in 2024 to 106 GW by 2035—more than a 200% increase over a decade—implying substantial capital expenditure and infrastructure buildout.

Electricity growth also appears across households and businesses

At the consumer level, AI does not replace electricity use but adds to it. Residential electricity demand in the United States is forecast at 1,535 TWh in 2026, up from 1,517 TWh in 2025. Commercial demand is projected to rise from 1,486 TWh to 1,525 TWh over the same period, indicating that electricity growth is coming from multiple sources, including AI and broader electrification.

Generation mix changes more slowly than demand

As demand grows faster than clean energy deployment, the system may rely longer on conventional generation to maintain stability. The share of renewable energy is forecast to increase from about 24% in 2025 to 27% in 2027, while natural gas remains around 40% in 2025–2026. That mismatch can increase emission pressures as the grid leans on dispatchable power.

Implications for the power system

Overall, AI is reshaping the energy system by changing how electricity demand scales with infrastructure expansion. As models grow larger and data centers expand, electricity demand does not rise linearly; it grows with the scale of physical buildout. If expansion outpaces the grid’s ability to adapt, the central question becomes whether there is enough electricity—and enough clean electricity—to sustain AI growth.

Source: Financial Times, Reuters