The most regressive tax in America right now is not levied by Congress. It is the surcharge quietly loaded onto household electricity bills to build the grid that AI data centers demand. Residential prices have climbed more than 36 percent since 2020, and one analysis pegs the ratepayer share of the $1.4 trillion buildout at roughly $700 billion. That is a wealth transfer from families to hyperscalers, and it should end.
Who pays for the machines that print money
Utilities plan to spend $1.4 trillion on AI-driven capacity, a 27 percent capex surge. Under standard rate design, those costs are smeared across the entire customer base, so a retiree in a two-bedroom apartment subsidizes the transmission upgrades that let a trillion-dollar company train models. A November 2025 survey found 78 percent of Americans worried their bills will rise because of data centers. They are right to worry, and right to be angry.
Large-load tariffs are necessary but insufficient
At least 36 utilities have adopted large-load tariffs, and the House Ratepayer Protection Act would push states toward a "large load standard" requiring data centers to fund their own grid upgrades. Good. But tariffs are riddled with exemptions, sweetheart economic-development deals, and cost allocations that still leak onto ordinary ratepayers.
- Ring-fence the costs: new interconnection, generation, and transmission for a data center should sit in a dedicated cost pool the operator pays in full.
- Take-or-pay contracts: operators forecasting enormous demand should commit to it financially, so ratepayers are not stranded if a project shrinks.
- Bring your own power: credit operators that add firm generation, and penalize those that consume capacity without adding any.
The voluntary pledges prove the point
Microsoft has pledged to cover its own electricity costs and Anthropic promised to absorb price increases tied to its buildout. If the largest players concede they should pay, the principle is settled. The only question is whether regulators codify it or let quieter operators keep free-riding. More than 300 data-center bills across 30-plus states show the politics have finally caught up.
This is a pricing failure, not an AI problem
None of this is anti-technology. AI infrastructure can pay for itself many times over; that is precisely why the companies building it can afford their own grid. The failure is regulatory inertia that defaults to socializing costs and privatizing gains. Fix the rate design and the market sorts the rest.
Electricity is not free, and pretending otherwise on the backs of households is bad economics and worse politics. Utilities have a further incentive to look the other way: under cost-of-service regulation, they earn a guaranteed return on every dollar of capital they deploy, so a $1.4 trillion buildout is not a burden to them but a bonanza, whoever ends up paying for it. That conflict is exactly why regulators, not utilities, must draw the line. Make the data centers pay for the power they demand. If a project only pencils out when strangers cover the wiring, it was never viable to begin with.