U.S. residential electricity prices have climbed 36% since 2020, rising from 12.76 cents per kilowatt-hour to 17.44 cents as of February 2026, with the U.S. Energy Information Administration projecting further increases to 19.01 cents by September 2027. Communities in data center corridors from rural Virginia to the Arizona desert have increasingly blamed AI hyperscalers — Microsoft, Google, Amazon, Anthropic, and OpenAI — for straining local power grids and driving up household energy bills. The backlash has reached the White House, where President Donald Trump acknowledged the industry "needs some PR help" and facilitated a Ratepayer Protection Pledge committing hyperscalers to absorb additional energy costs rather than pass them to consumers.
SemiAnalysis, a semiconductor and infrastructure research firm, puts most of the blame on an obscure market mechanism rather than data center demand alone. In the PJM Interconnection region — a grid serving 13 eastern U.S. states and home to major data center clusters — a pricing structure called the Base Residual Auction (BRA) requires utilities to pre-pay for anticipated electricity capacity roughly two years in advance. SemiAnalysis argues that PJM's proprietary demand forecasting models have systematically overestimated future load, partly because many planned data centers have faced construction delays tied to a <a href="/news/2026-03-14-tsmc-n3-wafer-shortage-ai-compute-2026">chronic memory component shortage</a>. Since capacity costs are socialized across all ratepayers regardless of actual usage, any demand overestimation creates a direct cost burden on households and small businesses. As a counterexample, the firm points to the Electric Reliability Council of Texas, which runs an energy-only market without a centralized capacity auction and has seen more stable prices since 2022 despite similar hyperscaler buildout — though that comparison carries limits, given the rate volatility Texas experienced in the wake of Winter Storm Uri in 2021.
Analysts have expressed skepticism about whether the hyperscalers' <a href="/news/2026-03-14-meta-weighs-20-workforce-layoffs-to-offset-rising-ai-infrastructure-costs">pledges to absorb costs</a> are financially credible. Marc Einstein, research director at Counterpoint Research, noted bluntly that "the industry's not making money, so that puts even more pressure on them." Maeghan Rouch, a partner at Bain and Company, acknowledged that constrained capacity markets like PJM have seen dramatic price increases tied to data center demand growth, while also cautioning that grid modernization spending and broader inflation contribute to rate increases independent of tech industry activity. Chris Howard, head of data centers account management at JLL, suggested that cost-absorption commitments could help companies secure community support for future projects, framing the pledges as strategically motivated as much as altruistic.
PJM's demand forecasting methodology is proprietary and not subject to independent public audit — a structural problem no corporate pledge can fix. State utility commissions in PJM's footprint, including those in Pennsylvania, Maryland, and New Jersey, have previously called for greater transparency in load forecasting, but any revision to BRA rules requires Federal Energy Regulatory Commission approval and would face opposition from merchant generators that benefit from elevated capacity payments. The incentive asymmetry is stark: overforecasting carries no institutional penalty for PJM, while ratepayers across 13 eastern states absorb the cost. Until that market design flaw is addressed, the bill for AI's electricity buildout will keep landing on households regardless of what hyperscalers pledge.