Who Pays for the Brain

In June 2025, electricity bills in New Jersey jumped roughly 20% overnight. No extreme weather. No fuel crisis. PJM — the grid operator covering 13 eastern states and 67 million people — held its annual capacity auction, and the price that had cleared at $29 per megawatt-day the year before cleared at $270. PJM's own analysis showed why: data center load growth. The AI buildout had consumed so much anticipated grid capacity that the auction had to reflect a system under real strain.

The families in New Jersey did not build the data centers. They do not own stock in the companies that did. Most of them have never thought about where the electricity goes when they ask ChatGPT a question. But they got the bill. This story is national: who finances the infrastructure of the AI era, and why the answer keeps coming out the same way — the people who live near it, not the companies building it.

The map

The K-Shape

The electricity market has produced its own K-shape. The sector driving demand is paying 3 cents more per kilowatt-hour. The sector with the flattest demand growth is paying 25% more over four years. The infrastructure cost flows downhill — toward whoever has the least power to negotiate it away.

The mechanism is simple. When a data center connects to the grid, the utility builds the transmission lines, substations, and generation capacity. Those infrastructure costs go into the utility's rate base — recovered from all customers, not just the one that triggered the need. A single hyperscale data center can require more new grid infrastructure than a small city. The residents nearby did not choose it. They pay for it through rates they cannot negotiate.

Oregon makes the disparity visible with a single number: over ten years, residential electricity rates climbed 8 cents per kilowatt-hour. Large commercial users — data centers among them — paid only 2 cents more. Hillsboro, a Portland suburb with 15 major data centers, saw nearly all of Portland General Electric's load growth come from commercial customers. The residential customers, who contributed almost none of that growth, paid four times the rate increase. The winter of 2025 brought a wave of electricity shutoffs during a cold snap — an affordability crisis directly connected to rates that had been rising while incomes had not.

Virginia: Where the Bills Are Already Filed

Virginia has nearly 600 data centers — more than any other place on earth. Northern Virginia's "Data Center Alley" routes approximately 70% of all global internet traffic. Data centers now consume 25–40% of the state's electricity, and demand is projected to rise 183% by 2040.

In November 2025, Dominion Energy won its first base rate increase since 1992. The state's utility regulator approved $565.7 million for 2026 — adding $11.24 per month to a typical residential bill. Average Virginia household electricity bills reached $167 per month in 2025, up 27% since 2021. The 33-year rate freeze ended because data center demand made it mathematically impossible to absorb the required infrastructure investment without raising rates.

Virginia responded with a new tool: the GS-5 rate class, effective January 2027, requires data center customers using over 25 megawatts to sign 14-year contracts and pay minimums regardless of actual usage. The idea is to prevent cost-shifting to residents going forward. Whether it works depends on whether the contracts hold — and whether 14 years of AI infrastructure commitment is even a realistic planning horizon for companies whose product cycles run in months.

The Texas Exception

Texas is the second-largest data center state with 9.7 GW of demand. Its residential electricity rates have been roughly stable for three years. The reason is ERCOT — Texas's deregulated competitive electricity market, where consumers can choose from multiple providers and switch when rates rise. There is no captive rate base into which infrastructure costs quietly flow and get recovered from customers with no alternative. The cost pressure that concentrates in PJM states dissipates differently in a competitive market.

This is not a case for deregulating electricity everywhere. The 2021 Texas freeze killed hundreds of people and exposed how badly ERCOT's model can fail under stress. The rate stability data is real regardless — and what it points at is that the mechanism causing residential cost-shifting in PJM states is a regulatory design choice, not an inevitable consequence of building data centers. Some states are now trying to change that design.

What Changes — and What Probably Doesn't

In March 2026, tech executives signed a White House pledge to limit their impact on residential electricity bills. It is a political document. There is no enforcement mechanism, no penalty for non-compliance. Meta's Hyperion campus in rural Louisiana just expanded from 2,250 acres to 3,650 acres — twice the size of the state's international airport. The companies signed. They are also still building.

Oregon's POWER Act is more serious: it requires data centers to pay for the actual strain they place on the grid, rather than pooling those costs across all ratepayers. If it holds up and produces measurable results, other states have a template. Community opposition has already blocked $98 billion in projects since 2025 — including Google's $1 billion Indianapolis project, pulled after public resistance. Maine passed a statewide moratorium on new data centers. None of this changes the rate structure. But it slows the pipeline, and slower pipeline means slower rate pressure.

EIA forecasts residential electricity prices could rise up to 40% by 2030 versus today. Data centers are projected to consume 12% of all US electricity by then — up from 4.4% today. The question the next four years will answer is whether the cost allocation mechanism gets redesigned before that demand arrives, or after.

The Read

EP18 said the war damaged the economy's fuel line but did not touch its brain. EP22 said Intel is the spine of that brain. The question this week is who pays for the electricity that keeps the brain alive — and the answer, right now, is people who live within a few miles of a data center and have no practical ability to avoid the cost.

The free market argument for AI data centers is that they generate growth, jobs, and capability that benefits everyone. That argument is real. The problem is that the benefits are diffuse — spread broadly across users and companies over time — and the costs are concentrated. The electricity cost accrues immediately and in full to whoever happens to live near the server farms. John Steinbach in Manassas got a $281 electricity bill in January when his typical bill was $100. He has never worked in tech. He did not vote on whether a data center should be built nearby. The market that produced this outcome is functioning exactly as designed. The question is whether it was designed the right way.

The companies building AI are making the most consequential infrastructure investment in a generation. The people financing a meaningful share of that infrastructure are John Steinbach and 67 million others in the PJM grid who did not choose this and cannot opt out. That is the deal the regulatory structure created. Some states are now trying to renegotiate it. ~ Framework