Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

Who Pays for AI’s Electricity? The Grid Cost Fight

The first real governance battle of the abundance era isn’t about robots taking jobs. It’s about whose electric bill goes up to train the robots.

Open your power bill. If you live in the thirteen states served by PJM, the largest grid operator in North America, somewhere inside that number you are quietly paying to train an AI model you will never use, owned by a company whose market cap exceeds your state’s GDP.

That isn’t a metaphor. PJM runs an annual auction to line up enough generating capacity to keep the lights on three years out. The auction for the 2025/2026 delivery year cleared at $269.92 per megawatt-day, an 833% jump over the prior year’s $28.92. The grid’s own Independent Market Monitor traced 63% of that increase directly to data center demand, which works out to about $9.3 billion in extra costs spread across ordinary households and businesses. The next auction cleared higher still, slamming into the FERC-approved price cap at roughly $329 per megawatt-day across the entire footprint. The NRDC projects $100 to $163 billion in cumulative cost through 2033 if nothing changes. For 2027/2028, nearly all of the forecast jump in peak demand, about 5,100 of 5,250 megawatts, is data centers.

Here is the part that should make you put the bill down. None of those households voted for an AI buildout. They are subsidizing it anyway, because of a century-old accounting habit nobody ever stopped to question.

The Default Nobody Chose: Socialize the Cost

For a hundred years, electric utilities have operated as regulated monopolies on a simple bargain: build the wires and the plants, then spread the cost across everyone who plugs in. When a new factory or a new suburb arrived, the grid expanded and the bill was shared. The logic held because demand grew slowly and predictably, and because a new factory created local jobs that justified the shared investment.

Then a single hyperscale data center started asking for a gigawatt, the draw of a mid-sized city, on a two-year timeline, while employing a few dozen people. The old bargain quietly broke. Spreading that cost across all ratepayers means a retiree in Ohio pays a little more so that a trillion-dollar company can run the GPU warehouses training frontier models. The cost is socialized. The benefit, as we argue in the piece on why public AI ownership still isn’t access, is privatized straight to shareholders.

This is the move that defines the early transition, and almost nobody noticed it happening because it arrived disguised as a line item. Socializing infrastructure cost is invisible by design. You don’t get a notice that says “your rate went up to power Azure.” You just pay 8% more and assume it’s inflation.

The Correction: Make the Beneficiary Pay

The backlash, when it came, was bipartisan and fast, because nothing unites a Republican from Missouri and a Democrat from Connecticut like a constituent’s power bill going up to subsidize Big Tech.

Oregon moved first. The POWER Act (Protecting Oregonians With Energy Responsibility), signed in 2025, ordered state regulators to carve out a separate rate class for “large energy users” drawing 20 megawatts or more, mostly data centers and crypto miners, so that the grid costs they cause are billed to them rather than smeared across families. It also requires those users to sign power-purchase agreements of at least ten years, committing them to pay for the projected demand and any new transmission their arrival makes necessary. In May 2026, Oregon regulators approved the pricing framework for Portland General Electric, and the new rates took effect. Oregon became the first state to formally separate the people causing the cost from the people paying it.

Washington at the federal level produced a flurry. The GRID Act (Guaranteeing Rate Insulation from Data Centers Act), introduced in February 2026 by Senators Hawley and Blumenthal, would push data centers onto power sources separate from the shared grid entirely. The House has the “Protecting Families from AI Data Center Energy Costs Act” and a draft Ratepayer Protection Act giving states a template to shield consumers. The administration convened seven leading AI companies in March 2026 to sign a voluntary Ratepayer Protection Pledge, committing them to build, bring, or buy enough new generation to cover their own demand and to pay the full cost of the delivery infrastructure they require. Sanders and Ocasio-Cortez went further and proposed pausing new construction outright. Across more than thirty states, lawmakers filed 300-plus bills on data centers in 2026 alone.

Strip away the partisan packaging and every one of these proposals encodes the same principle: the entity that captures the value should carry the cost. Economists call it internalizing the externality. Your grandmother called it cleaning up your own mess.

The Third Path: Build Your Own Grid

While legislators argued, the hyperscalers did what the rich always do when a shared system gets expensive. They started building their own.

Behind-the-meter power, on-site generation that never touches the public grid, has gone from workaround to strategy. A modern data center campus increasingly looks like a private utility: gas turbines, battery banks, solar arrays, sometimes a contracted small modular reactor, wired into a microgrid the operator owns outright. xAI’s Colossus facility in Memphis runs on on-site gas turbines that bypass the grid; the supply problem behind that scramble is the subject of the Electron Gap. The hyperscalers, as one grid executive put it, didn’t set out to become power companies. The grid’s wait times left them no choice. Google has even begun pairing with virtual power plants, networks of distributed batteries and flexible demand, to free up capacity instead of waiting years for new wires.

On paper this solves the cost-allocation fight by making it moot. If a data center generates its own power, it can’t shift costs onto your bill. But notice what it actually is: grid defection by the wealthiest actors in the economy. When the strongest customers exit the shared system, the fixed cost of maintaining that system, the poles and substations and the people who repair them after storms, falls on a smaller and poorer pool of everyone left behind. The book’s chapter on Forking treats exit as a healthy check on captured institutions. But exit by the powerful, leaving the powerless to hold the bag, is the failure mode, not the feature.

Why This Is the Book’s Fight, Not a Footnote

It would be easy to file all this under “boring utility regulation.” That would be a mistake. The cost-allocation question is the first concrete, dollars-and-cents instance of the choice the entire Unscarcity framework is built around: when a technology starts producing abundance, who captures the surplus and who eats the cost?

Get this wrong and you don’t get a post-scarcity civilization. You get a feudal one with better chips, where a handful of compute landlords run private power empires and externalize every cost they can onto a public grid they’ve already abandoned. That is the Star Wars timeline the Preamble warns about: abundance technology captured as private rent.

The Foundation treats energy as a birthright, part of the survival floor that should be guaranteed to everyone. But a birthright has to be funded, and the funding principle is the whole ballgame. The Foundation works only if the cost of the surplus-generating machinery is borne by the surplus it generates, not by the households the surplus is supposed to serve. That is exactly the logic of Land Tax Funds Abundance: capture value where it actually accrues. Oregon’s rate class is a crude, early, real-world version of the same idea. Make the beneficiary pay, and the abundance can be shared. Socialize the cost while privatizing the gain, and you have simply invented a new scarcity and handed the bill to the people with the least.

There is a deeper reason energy sits at the center of this. If labor stops being the thing we trade and kilowatt-hours become the unit that backs value, then how a society allocates the cost of generating those kilowatt-hours becomes the master question of its political economy, the way control of land was for the agrarian age and oil for the industrial one. The grid is where the abundance transition becomes physical, and metered, and impossible to hand-wave.

The Honest Bottom Line

We are running the experiment right now, in public, on the PJM grid and in the Oregon statehouse, and the early returns are clear enough to act on.

Socializing the cost is the path of least resistance and the worst outcome: invisible, regressive, and a direct subsidy from the many to the few. Internalizing the cost, the Oregon and GRID Act direction, is harder and slower and exactly right, because it forces the beneficiary to carry the infrastructure it demands. The behind-the-meter third path can help, if it adds genuinely new generation, or it can become grid defection that strands everyone else, depending entirely on whether the rules make the builders pay their share of the commons they’re leaving.

The technology that closes the energy gap may well be fusion arriving in the 2030s. But fusion is a supply answer to a demand problem, and it does nothing about who pays in the fifteen years before it shows up. That question is being decided now, in rate cases and committee markups most people will never read, and it will quietly determine whether the AI buildout produces shared abundance or the most efficient upward wealth transfer in history.

Your power bill already cast a vote. The only question is whether anyone asked you first.

Who Pays for AI's Electricity? The Grid Cost Fight