Who Pays for AI Infrastructure — and Who Gets to Decide?

Here is something worth pausing on: one state has started to answer, in law, the question every data-center fight ends on.

Not whether the technology is good or bad. Who pays.

Minnesota created a ‘very large’ customer class — in practice, data centers — and now requires regulators to structure service so that costs attributable to those customers stay with that class, with protections against leaving ordinary ratepayers with stranded costs. And there is an annual fee, two to five million dollars, on the largest sites.

That is real, and Minnesota has put some of those protections directly into state law. It looks like an answer to the question.

But look at what the same legislature did not do, and the answer falls apart.

Minnesota answered who pays. It has not answered who decides. Those are not the same question — and mistaking one for the other is the mistake that keeps these fights going in circles.

---

Two Questions, Not One

Who pays is a question about cost. It can be priced. A full-cost tariff, a minimum payment, an exit fee — these are instruments for making sure the new customer bears the burden it creates. Minnesota built those instruments.

Who decides is a question about power. It asks whether the people who will live beside the project have any standing over whether it is built, where it goes, and on what terms. That cannot be priced. No tariff answers it.

The distinction is not subtle. A community can be fully protected from subsidizing a data center — and still have no say over the fact that it is being built across the road.

---

What Minnesota Did Not Pass

The same legislature that wrote the cost protections considered, and failed to pass, three bills aimed squarely at the second question.

A bill to ban the non-disclosure agreements developers use to hide projects from residents until the plans are effectively locked. It passed one committee unanimously, then died in another, after business groups spoke against it.

A bill to require a full environmental review of every data center. It never became law.

A bill to pause new permits until the state actually studied what these facilities do to energy, water, and communities. It never got a hearing.

So the state moved directly on who bears electrical costs, while proposals that would have strengthened transparency, environmental review, and the public’s position before projects advanced did not become law.

The gentler version of the same story is the work the state does fund. The University of Minnesota is building a decision-making toolkit for small towns weighing a proposal, and its researchers are careful to say the job is not to tell a community whether to accept or reject. It is to give better information.

That is valuable. But information is not authority. Knowing which questions to ask is not the same as having the power to act on the answers. And the gap is not abstract: when the state’s Department of Natural Resources reviewed one proposed facility’s environmental analysis, it found gaps and flaws in the review itself — in a state where no law requires a full review in the first place.

---

The Other End of the Same Question

The contrast has a name: Memphis.

When xAI announced what it called the world’s largest supercomputer there in 2024, the public had effectively no input into the decision. The negotiations had happened behind closed doors. The gas turbines went in without permits — residents raised the alarm before the permitting system did. The benefits that followed were real: school funding, a wastewater-recycling facility, a share of tax revenue aimed at nearby neighborhoods. But they arrived after the decision, and nothing retroactively created the standing that was missing when the bargain was struck.

Minnesota and Memphis look like opposites, and in one narrow way they are. Minnesota legislated the cost question before the projects landed; Memphis let the projects land first and built participation afterward.

But on the second question — who decides — they are closer than either side would like. Neither has answered it. Minnesota protected the ratepayer’s wallet and left the resident’s voice unresolved. Memphis settled neither.

---

The Costs Are Getting Standardized. The Voice Is Not.

Underneath all of this runs a larger movement, and it points one way.

Utilities and regulators are converging on the same answer to who pays. By one count, more than sixty large-load tariffs are now in force or pending across thirty-four states — fifteen-year contracts, minimum payments, collateral, exit fees. The machinery for making a big new customer pay its own way is becoming routine.

What is far less standardized is the answer to who decides: how residents gain meaningful standing over siting, secrecy, environmental review, and the local consequences of a facility. The cost question is increasingly being standardized. The power question is not.

Even Minnesota’s answer to the cost question should be held carefully. On May 15, 2026, the Minnesota Public Utilities Commission approved Xcel Energy’s large-load tariff — a standalone customer class for loads at or above 100 megawatts, fifteen-year default contracts, an 80 percent minimum bill, exit provisions, and customer-funded upgrades. The design is no longer merely statutory; it has entered regulatory implementation. Whether it survives contact with an operating project’s economics is still an open question.

---

The Question That Remains

None of this is an argument for data centers or against them. The same building reads as a paycheck to one worker and a threat to another — construction work on one side of the table, automation anxiety on the other. The point is not that anyone is wrong. It is that the two questions have to be kept separate, because answering the first tells you nothing about the second.

The ledger still applies. What is actually true. Who bears the cost. Who receives the benefit. And then the fourth question, the one that keeps getting skipped: what would a fair decision process look like?

Minnesota is the clearest evidence yet that the first three, answered well, do not answer the fourth. You can get the cost allocation exactly right and still be standing in front of a room of people who were never given a seat where the decision was made.

That is the distinction to carry into every one of these fights.

Cost-bearing is not standing. A community can be protected from paying for AI infrastructure and still have no power over whether, where, and how it is built. Designing the first is necessary. Designing the second is the part almost no one has begun.

---

Dale Joseph is the author of Thought Partners: Preserving Cognitive Sovereignty in the Age of AI and founder of the Emergence Institute. He worked for years as a consultant helping install hospital networks before turning to writing and systems thinking. He lives in Boynton Beach, Florida.