AI's Power Surge Hits US Electric Bills

Power prices on the largest electric grid in the US jumped 76% in the first quarter due to rampant demand from data centers, with wholesale power on the 13-state PJM Interconnection LLC grid averaging $136.53 per megawatt-hour in the first three months of the year, according to Bloomberg reporting on a Monitoring Analytics report . That compares to $77.78 per megawatt-hour during the same period in 2025 .

The report from Monitoring Analytics, PJM's independent market monitor, states bluntly that "data center load growth is the primary reason for recent and expected capacity market conditions" and warns that "the current supply of capacity in PJM is not adequate to meet the demand from large data center loads and will not be adequate in the foreseeable future" . The PJM grid serves 67 million people across 13 states including Virginia, Pennsylvania, Ohio, and Illinois—nearly 20% of the US population.

The price spike isn't just a number on a spreadsheet. John Steinbach, a Virginia resident, was shocked to receive a $281 electricity bill in January 2026—a huge spike from the roughly $100 he'd paid the previous month, according to Consumer Reports . Nearly three-quarters of Virginia voters blame data centers for rising electricity costs, according to a January 2026 survey by Global Strategy Group and the Chesapeake Climate Action Network Action Fund .

Can Grids Keep Pace With AI's Appetite?

The numbers behind the demand surge are staggering. The Lawrence Berkeley National Laboratory predicts that US data center demand will grow from 176 terawatt hours in 2023 (about 4.4% of total US electricity consumption) to between 325-580 TWh (6.7-12.0%) by 2028, according to a Belfer Center report . Morgan Stanley Research forecasts US data center demand could reach 74 GW by 2028, with a projected shortfall of about 49 GW in available power access .

The International Energy Agency projects global electricity consumption for data centers will double to reach around 945 TWh by 2030 in its Base Case, representing just under 3% of total global electricity consumption . Electricity consumption in accelerated servers, mainly driven by AI adoption, is projected to grow by 30% annually in the IEA's Base Case .

The infrastructure challenge extends beyond generation. In traditional legacy hubs such as Frankfurt, London, Amsterdam, Paris, and Dublin, connecting data centers to the grid typically takes 7 to 10 years, with some delays up to 13 years, according to Eurelectric citing an Ember report, while the IEA notes that in advanced economies it typically takes four to eight years to build high-voltage lines .

What's China Doing Differently?

While US grids strain under AI demand, China is pioneering a novel approach. China's large-scale data centers have joined electricity spot trading as virtual power plants for the first time, with three data center clusters operated by China United Network Communications Group and China Mobile in Guangdong entering the spot market on May 14 via a Guangdong Power Grid platform, Bloomberg reported . The system allows data centers to buy electricity based on real-time prices .

The China Unicom Greater Bay Area Data Center has transitioned from a "major electricity consumer" to a "flexible resource" role, according to China Money Network . This marks a significant shift in how computing demand interacts with the grid, allowing facilities to adjust power consumption based on price signals and grid conditions.

China is also directly linking renewable energy to data centers. China's first large-scale project to supply renewable energy directly to data centers started operating in early May in Ningxia, with a solar capacity of 500 megawatts and a further 1.5 gigawatts of wind power planned by year-end, according to the South China Morning Post . The project is designed to coordinate energy supply and computing capacity so the data center can shift tasks to where renewable power supplies are plentiful or when electricity is cheapest .

Data centers in China accounted for 1.2% of total power demand last year, posted a 38% compound annual growth rate over the past five years, and are forecast to maintain a 19% CAGR through 2030, according to Rystad Energy .

Who Pays for the Buildout?

The cost allocation question is becoming politically explosive. In recent PJM capacity auctions, data centers accounted for roughly 63% of the price jump in one key auction, translating to an estimated $9.3 billion in extra costs passed to customers across the 13-state grid over the next year, according to Energy News Beat .

A September 2025 analysis by the Union of Concerned Scientists found that in 2024 alone, utilities in seven PJM states approved $4.356 billion in local transmission projects needed solely to connect data centers to the grid, with over 95% of these costs socialized across all utility customers rather than assigned to the data centers causing them, and from 2022-2024 the total exceeded $5 billion .

The US Energy Information Administration projects average residential electricity prices will rise 5.1% in 2026, and PowerLines estimates that residential customers could bear approximately $700 billion of a projected $1.4 trillion in utility spending through 2030, with consumer electricity bills having already risen approximately 40% since 2021 .

Consumer advocate Cathy Kunkel told NPR that "I think it's almost inevitable, the way that these structures are set up, that ordinary people are going to end up subsidizing the wealthiest industry in the world" unless reforms ensure more costs are paid by data centers themselves .

Can Chips Get More Efficient?

As power constraints threaten to limit AI scaling, semiconductor energy efficiency has become critical. Since 2010, overall semiconductor application energy use has doubled every three years, and by 2030, if this rate continues, the semiconductor industry could consume nearly 20% of global energy production, according to the US Department of Energy .

With AI data center workloads forecast to triple or quadruple annually between 2026 and 2030, chip manufacturers in 2026 are increasingly integrating high-bandwidth memory closer to logic chiplets, allowing data to move much faster between processors and memory at multiple terabytes per second while being more energy efficient, according to Deloitte's 2026 Semiconductor Industry Outlook .

In a breakthrough published in Science in early May, a UC Berkeley-led team discovered they can transform titanium dioxide into a ferroelectric material by reducing its thickness to less than 3 nanometers, potentially opening a pathway toward ultra-scaled, energy-efficient electronic devices .

The buying conversation is already shifting from peak compute performance to FLOPS-per-watt and cost-per-inference, with software-level inference optimization—model distillation, quantization, compiler tuning—becoming where meaningful performance gains come from in 2026 and 2027, according to semiconductor experts at HTEC .

What Changed This Week

The 76% power price jump on PJM, reported this week by Bloomberg, crystallizes how AI's infrastructure demands are translating into household costs faster than policymakers anticipated. China's launch of spot trading for data centers as virtual power plants on May 14 demonstrates an alternative model where computing facilities become flexible grid resources rather than inflexible loads. Meanwhile, the semiconductor industry is racing to improve chip efficiency as energy supply emerges as the binding constraint on AI scaling, with breakthrough materials research from UC Berkeley offering potential pathways forward.

What to Watch

The US Department of Energy has identified 16 potential sites for AI infrastructure development with in-place energy infrastructure and ability to fast-track permitting for new energy generation such as nuclear, with an RFI released this week seeking input from developers . Maryland's Office of People's Counsel has filed a complaint at FERC challenging PJM cost allocation rules, arguing Maryland customers could be assigned $2 billion in transmission capital costs tied to data center-driven upgrades, raising bills by $1.6 billion over the next decade . In 2026 so far, lawmakers in more than 30 states have introduced over 300 bills on issues related to data centers, including moratoriums, tax incentives, and energy policy, according to Multistate . Watch for FERC rulings on cost allocation, state-level data center legislation, and whether hyperscalers' pledges to bring their own power generation materialize at scale.

According to market data, WTI crude traded at $71.50 per barrel on Friday, up 0.6%, while Brent crude stood at $75.20 per barrel, up 0.5%. Henry Hub natural gas fell 2.4% to $3.25 per MMBtu, reflecting continued pressure on gas prices even as utilities plan massive buildouts of gas generation to serve data centers.