AI Data Centers Are Turning Power Bills Into Political Invoices
Reuters reports that a growing number of U.S. states are letting utilities charge customers for power-plant and grid-construction costs before those projects are finished, a financing mechanism known as construction work in progress, or CWIP. The number of states using some version of the tool has roughly doubled over the past decade to about 40. Utilities say the policy helps them build capacity fast enough for rising electricity demand, including the AI data-center boom. Industrial and consumer advocates warn it moves risk from shareholders and large customers onto ordinary ratepayers before the promised benefits arrive. The political issue is simple: AI may be sold as software, but it lands as steel, concrete, land, water, transmission lines and higher monthly bills. When public regulators let monopoly utilities recover costs early, voters become silent lenders to the cloud economy.
The important thing about the AI boom is that it keeps escaping the place where people were told it lived. It was supposed to be a software story. Then it became a chip story. Then a data-center story. Now Reuters is showing the next layer: it is becoming a utility-bill story. That is where the politics get real.
A model running in a server farm does not feel virtual to the family opening a higher electric bill. A cloud contract does not feel innovative to the small manufacturer paying more for power while the largest technology companies negotiate special terms. The story underneath the headline is not whether AI is useful. Of course it is useful. The story is who gets the upside, who absorbs the risk, and who discovers too late that “infrastructure investment” often means the public balance sheet was quietly drafted into service.
Construction-work-in-progress financing sounds like a procedural utility-policy term. That is why it matters. The most consequential transfers usually arrive in boring language. If a state lets a utility charge customers before a plant or grid upgrade is finished, the customer is no longer merely buying electricity. The customer is helping finance construction. If the project comes in late, costs more than expected, or serves a narrow group of giant loads, the risk does not stay neatly inside a corporate spreadsheet. It gets socialized through rates.
This is the second-order effect that the AI debate keeps missing. The public argument is full of demos, job forecasts, safety pledges, and stock-market excitement. But the bottleneck is physical. Compute needs power. Power needs wires, substations, generation, permitting, water, land, and political permission. Every one of those bottlenecks creates a negotiation between concentrated beneficiaries and dispersed citizens. The companies know exactly what they need. The utilities know exactly how to get guaranteed returns. The average household learns about the deal only after it shows up as another fixed cost of modern life.
That is not a left-right story. It is an incentive story. Regulators want capacity. Governors want investment headlines. Utilities want rate-base growth. Technology companies want speed and reliability. Local governments want tax base. Everyone at the table has a reason to say yes quickly. The people paying in small monthly increments are rarely organized enough to say, “wait, why am I underwriting this?”
The danger is not that America builds too much infrastructure. The danger is that it builds infrastructure through a political accounting trick that hides the true beneficiary. If an AI campus requires a dedicated generation buildout, then the cost should be transparent. If the public is subsidizing it, say so. If ordinary customers are being asked to front the capital, say so. If the project genuinely strengthens the grid for everyone, prove it before the charge appears on the bill.
There is a version of the AI buildout that could make the country more productive. There is also a version that looks like a private gold rush with public-rate financing. The difference will not be decided in a keynote. It will be decided inside utility commissions, tariff filings, interconnection queues, and statehouses where most voters are not watching.
That is why this belongs in political news. The next phase of AI will not only be argued over copyright, bias, chatbots, or jobs. It will be argued over who owns the physical layer beneath the digital economy. If politicians want to invite the data-center boom, fine. But they should be forced to put the invoice where voters can see it. The public should not find out after the fact that the cloud came with a line item on the power bill.