You May Pay for the AI Power Boom Before the Plants Even Work
Reuters reports that more states are letting utilities bill customers for large electricity projects before those projects are finished, reviving construction-work-in-progress policies that can shift delay and cost-overrun risk onto ratepayers. The mechanism is not abstract. Customers have already paid billions toward projects such as Georgia’s Vogtle nuclear reactors, a Nevada transmission project, and Virginia offshore wind development before full operation. Missouri recently reversed a decades-old ban on the practice, while Arkansas, Kansas, Oklahoma, and North Carolina have adopted similar provisions. The timing matters because the grid is being asked to absorb a new wave of large-load demand from data centers, AI computing, manufacturing, and electrification. In parallel, Reuters reported Hut 8 signed a roughly $9.8 billion, 15-year lease for a Texas AI data-center project, with a 352-megawatt first phase and a planned one-gigawatt campus. The public argument is reliability and investment. The citizen question is simpler: who pays when the political class promises a bigger grid before proving the project can be built on time and on budget?
The AI boom is usually sold as something weightless. Software. Models. Chips. Productivity. The future arriving through a prompt box.
Then the electricity bill shows up.
That is the part of the story voters are rarely asked to price. Reuters’ reporting on construction-work-in-progress charges gets at the real political economy underneath the boom. Utilities and state governments are increasingly willing to let customers pay for power projects before those projects are finished. In plain English: the risk moves from the company building the asset to the household or business that needs electricity.
That might be defensible in narrow cases. The grid does need investment. Delayed infrastructure can be more expensive than early infrastructure. A country that wants factories, data centers, and reliable power cannot pretend transmission lines and generation appear by magic.
But the public deserves to know when “investment” has become a pre-approved invoice.
This is where the AI story stops being a Silicon Valley story and becomes a local politics story. Data centers do not run on vibes. They need land, water, transformers, permits, gas, nuclear, transmission, and long-term power contracts. Hut 8’s Texas lease shows the scale: billions of dollars, hundreds of megawatts in the first phase, and a planned campus large enough to make electricity the central constraint rather than an afterthought.
Markets are already treating power as the scarce layer. Politicians are just slower to say the quiet part out loud: if the next technology boom needs a bigger grid, somebody has to finance that grid before the returns are obvious.
The danger is not that America builds infrastructure. The danger is that the cost allocation is decided before citizens understand the bargain. If a utility gets to earn returns while a project is unfinished, ratepayers are no longer just buying electricity. They are financing construction risk. If the project is delayed, they still pay. If demand forecasts are wrong, they still pay. If the data-center boom becomes more concentrated than promised, they may have subsidized a private buildout while being told it was a public necessity.
That is the incentive problem. Politicians get to announce investment. Utilities get regulatory certainty. Large power users get a path to capacity. Investors get a more bankable project. The dispersed household customer gets a line item.
None of this means every grid project is bad. It means the politics should be honest. If AI demand is creating a new infrastructure cycle, then the public debate should be about who captures the upside and who is forced to absorb the downside. A data center can sign a lease. A homeowner cannot easily opt out of the regulated utility bill.
The old internet trained people to think digital progress was almost free at the margin. The new AI stack is teaching the opposite lesson. Intelligence may scale in the cloud, but the cloud is built on physical bottlenecks. Power is the balance sheet. The grid is the gatekeeper. Rate regulation is where the private boom meets the public wallet.
That is why this story matters more than another round of AI hype. The real revolution is not a chatbot answering faster. It is the quiet redesign of utility finance so the country can build enough power for the machine economy. Maybe that buildout is necessary. But necessity is not a blank check.
If policymakers want citizens to fund unfinished projects, they should attach hard protections: transparent cost caps, clear large-load contributions, public reporting on overruns, and a real separation between ordinary reliability spending and speculative demand from private computing campuses.
Otherwise the AI boom will arrive with a familiar political trick. The gains will be marketed as innovation. The risks will be socialized as infrastructure. And the bill will be mailed to people who were never invited into the deal.