Goldman Sachs published data in February showing that electricity prices in the United States rose 6.9% in 2025 — more than double the headline inflation rate of 2.9% — and project an additional 6% increase through 2027. The proximate cause is data center demand, which now accounts for 40% of US electricity demand growth. The PJM Interconnection, the largest grid operator in the country, projects a six-gigawatt reliability shortfall by 2027 as load growth accelerates faster than generation can be brought online. The numbers have a political charge to them: every household in America is paying more for electricity, and a significant share of that increase is attributable to infrastructure that benefits primarily a handful of technology companies.
The question of who should pay for this is being answered, for now, by the grid operators' rate structures — and the answer is: everyone. A Washington Post investigation found that Silicon Valley companies have been quietly building a shadow power grid across the United States, negotiating direct contracts with power generators that bypass the regulated utility system and guarantee them priority access to electricity. This structure has two effects. It insulates the technology companies from the most acute supply shortages and price volatility. And it shifts the infrastructure costs — grid upgrades, transmission buildout, reliability investments — onto the ratepayers who remain inside the regulated system, predominantly residential and small commercial customers who do not have the negotiating leverage to go around it.
A December 2025 capacity auction run by PJM illustrates the scale. Data centers accounted for $23.1 billion of the $47.2 billion total cost of three consecutive capacity auctions held since mid-2024 — nearly half. Those costs are socialised across all customers on the grid, regardless of whether they consume AI inference or ever interact with a product whose model is trained at one of the facilities driving the demand. The CNBC analysis from earlier this month framed this as a ratepayer protection problem that is generating genuine political backlash, with strange-bedfellows coalitions of consumer advocates, environmental groups, and some state-level Republicans arguing that the current cost allocation is unfair.
The technology companies would argue — with some justice — that the same grid infrastructure they are financing is what enables the clean energy transition, since large-scale power purchase agreements drive the construction of new renewable generation. This is partly true. Microsoft, Google, and Amazon have all made commitments to match their consumption with new renewable generation, and the data center buildout has been a significant driver of utility-scale solar and wind investment. The grid upgrades being funded partly by data center demand are also upgrades that benefit the entire system. Whether these benefits are commensurate with the costs, and whether the cost allocation is equitable, are separate questions from whether the buildout is good or bad in aggregate.
What makes the current situation particularly fraught is the scale mismatch. Electricity infrastructure investments are made in thirty-year time horizons. The AI demand forecasts that are currently driving those investments are based on assumptions about model scaling, enterprise adoption, and inference economics that are, at best, speculative at five years out. If the AI buildout proceeds as the most optimistic projections assume, the infrastructure will be needed and the ratepayer cross-subsidy will look like a reasonable price for transformative technology. If the buildout overshoots — as data center construction has before — consumers will have absorbed the cost of stranded assets that primarily enriched a small number of technology companies during a speculative peak.
The political moment is odd. The same administration proposing to preempt state AI regulation to remove friction from the industry's growth is also presiding over an electricity price environment that is generating real populist anger, including from its own voter base. Bernie Sanders and Ron DeSantis have both come out against the data center boom on energy cost grounds — a coalition that should cause some concern for an industry that has been comfortable operating without serious political opposition. The anger is not yet organised into effective policy pressure. But the numbers keep compounding: 6% electricity price growth on top of existing inflation, sustained for two more years, reaches voters in a way that abstractions about AI productivity gains do not.