by Barry Keating, Ph.D.
I want to start in New Carlisle, Indiana, a town of about eighteen hundred people in St. Joseph County, which recently found itself negotiating with the largest company on earth — and doing so, at first, in secret.
When Amazon’s $11-billion data center campus came to New Carlisle, members of the Town Council signed nondisclosure agreements. The agreements barred them from disclosing what they called “nonpublic information relating to Amazon.” The town’s residents, in other words, were not permitted to know the terms of a deal that would reshape their water system, their roads, and their tax base for a generation — until the deal was essentially done.
The reported terms, when they emerged, were these: a 10-year, 50-percent abatement on real property taxes, and an 85 percent exemption on enterprise technology equipment for 35 years. Thirty-five years. There are children in New Carlisle today who will be grandparents before that exemption expires. Alongside it, a $114-million water and sewer agreement and a $7-million contribution to roads and an interchange.
Now, I am not here to tell you that New Carlisle got a bad deal. I am here to tell you something more unsettling: For a long time, nobody in a position to judge was allowed to know whether it was a good deal or a bad one. And that is the subject of my talk: the complete contract.
I come at this as a Public Choice economist in the tradition of James Buchanan and Gordon Tullock. Public Choice begins with a simple, almost rude assumption: People in politics behave like people everywhere else. They pursue their own interests. A mayor wants to announce a jobs triumph before the next election. A company wants the lowest cost and the least scrutiny. A bureaucrat wants a bigger budget and a quieter life. None of them is paid, day to day, to maximize something called “the public interest.”
That sounds cynical. It is actually optimistic, because it tells us what to do. If you cannot rely on goodwill, you must rely on rules. And a contract between a municipality and a data center is the most important rule a town will write in a generation. A complete contract is one that answers, in advance, four questions: who pays for what, who bears the risk when things go wrong, how do we measure compliance and what happens at the end.
An incomplete contract is not a neutral document. It is an invitation. Tullock taught us about rent-seeking: the art of getting rich not by producing something, but by persuading government to give you something. Every gap in a contract is a future rent-seeking opportunity, and the party with the lawyers will find it. So let us walk through the four places where these contracts are most often left incomplete — power, water, noise and land — and see what completeness looks like.
First, power. A data center is, electrically speaking, a small city that never sleeps. A single large campus can draw several hundred megawatts — the equivalent of a few hundred thousand homes — and it wants that power on day one. The question is who pays for the wires, the substations and the generation to serve it. Left vague, the answer defaults to: everybody else, through their electric bills.
Ohio gave us the cautionary version of the good answer. Under a tariff approved by the state’s utility commission, large data centers must commit to paying for at least 85 percent of their contracted load, for eight years, with a three-year exit fee and full reimbursement if they cancel a project after the utility has built for them. Think about what that does. It means a company cannot reserve a gigawatt of capacity, watch the utility build a substation, and then walk away leaving ratepayers holding the bill. The risk sits with the party that created it.
Indiana produced a structurally elegant variant: in the AWS–NIPSCO arrangement, the data center load is served through a separate generation company, ring-fenced from the ordinary rate base, with roughly a billion dollars in bill credits over fifteen years flowing back. The principle is what matters: the load that causes the cost is the load that pays for it.
And Virginia gave us the warning about what happens when nobody writes the number down at all. The state’s data center sales-tax exemption was projected, back in 2008, to cost about $1.5 million a year. By fiscal year 2025 it cost $1.9 billion. That is not a rounding error; that is fiscal illusion — a benefit that grows in the dark because no one put a cap, a sunset, or a review date in the original deal. A complete contract has an expiration date. An incomplete one has a surprise.
Second, water. A data center that cools with water can drink like a town. And water contracts are where good intentions go to evaporate.
Goodyear, Arizona, is my favorite story, because it is a story of a second chance. Microsoft originally promised to pretreat its cooling water on site and return roughly five million gallons a day to the city’s surface-water plant, where it would become drinking water. It was a beautiful promise. It failed, almost immediately, over regulatory trouble. The city renegotiated: Microsoft would pay $36 million toward a $90 million expansion of the 157th Avenue treatment plant, post a $5 million assurance for the sewer line, fund a city coordinator, and — critically — air-cool its later buildings instead of evaporating water. Goodyear got real concessions. But notice the arithmetic: $36 million of a $90 million plant. The residents still carry the majority of a facility sized partly for Microsoft. Renegotiation after construction has begun is negotiation with one hand tied.
Compare Douglas County, Georgia, where Google’s cooling water is one hundred percent reclaimed water, against The Dalles, Oregon, where the volume guaranteed to Google was kept secret as a trade secret — roughly a third of the city’s supply — and the public was told, in effect, that it was none of their business. Or Botetourt County, Virginia, where the agreement reportedly scales from two million gallons a day on day one to eight million gallons a day on what the documents call “Day N” — a delightfully vague phrase for a very wet future — with the cost share described only as “some or all.”
A complete water clause does four things: it meters the water, it caps the volume, it prices the water at its full cost and it publishes the numbers. If a clause cannot survive sunlight, it should not survive the council vote.
Third, noise. Noise is the most human of these problems, because it is the one residents hear at two in the morning. And noise clauses fail in a specific, predictable way: They promise quiet without defining it.
Prince William County, Virginia, did the unglamorous work: an ordinance with enforceable decibel limits at the property line, measured on the C-weighted scale that captures the low-frequency hum data centers actually produce. Of 19 complaints investigated, none was found in violation — not because the ordinance was toothless, but because the standard was clear enough that operators complied. Hermantown, Minnesota, went further in a proposed agreement: a 47-decibel cap, strict liability for well contamination, no tax abatement at all and a $210 million inflation-indexed parent-company guarantee for decommissioning. That is what completeness looks like at the end of a facility’s life: someone has already posted the money to clean up.
The lesson is simple. “The operator shall minimize noise” is not a clause; it is a wish. A clause names the number, names the meter, names who measures, and names the fine.
Fourth, land and taxes — where New Carlisle returns to the story. A tax abatement is a decision to let one taxpayer pay less for decades, which means everyone else pays more, or gets less, for decades. That decision deserves the same seriousness as a decision to build a school.
New Carlisle’s 35-year equipment exemption is the extreme case. But the catalogue of better practice is encouraging. La Porte, Indiana, renegotiated a Microsoft deal in March 2026 from an exemption-plus-payment structure to full property taxes, split 85 percent to the city and 15 percent to the school corporation over 20 years. A renegotiation, yes — but in the public’s favor, and it proves these deals are not carved in stone. In Decatur Township, near Indianapolis, the Sabey development paired roughly $240 million in abatements with a guaranteed minimum of about $30 million in community payments plus at least $53 million in taxes during the abatement period. A floor, not a hope. And in Tucson, Arizona, we saw the opposite failure: The city unanimously rejected a 290-acre, 600-megawatt campus called Project Blue, but the county had already rezoned the land and sold it, and the power route through the state regulator stayed open. A city “no” that does not stop a project is a lesson in fragmented jurisdiction — the developer simply shopped for the friendliest government in the pile.
So: price the load, meter the water, name the decibel, floor the taxes, bond the cleanup, and — above all — write it down where the public can read it before the vote, not after.
Now, you may be thinking: This sounds expensive, and slow, and lawyers will eat the town’s budget. Let me offer the counterexample that convinced me completeness is practical. It is an Indiana story, and it is about a road.
In 2006, Indiana leased the 156-mile Indiana Toll Road to a private concession company for 75 years, in exchange for $3.85 billion up front. The money funded Major Moves, the state’s highway program, paid off the road’s debt, and left a trust fund. But the interesting part is not the money. It is the contract. The lease capped how much tolls could rise. It required the operator to fund repairs, maintenance and expansion. It set cleanliness and emergency-response standards. It required the operator to expand the road if congestion hit specified triggers. It barred the state from building a competing four-lane highway within 10 miles on either side — a non-compete clause that protected the operator’s investment, honestly bargained. It made the operator reimburse the state for policing and for monitoring the contract. And it fined the operator for underperformance.
Here is the punchline, and it is the best argument for completeness I know. The operator went bankrupt. In 2014, done in by the Great Recession and traffic projections that never came true, the concession company filed for Chapter 11. And nothing bad happened to the public. There was no interruption in service. There were no toll increases caused by the bankruptcy. There was no taxpayer bailout. The $458 million the operator had invested in road, bridge and pavement improvements stayed in the road. A new operator took over the remaining term.
Why? Because the contract had decided, eight years earlier, who would lose if the traffic forecasts were wrong. The answer was: the people who made the bet. Not the drivers. Not the taxpayers. Completeness does not mean predicting the future. No contract can do that. Completeness means deciding, in advance, who bears the cost when the future surprises you. That is the whole ballgame.
So let me close with a checklist for your next council meeting, committee or negotiation. A complete data center contract:
- Names the power load, assigns the grid-upgrade costs to the load that causes them, and sets minimum bills and exit fees.
- Meters, caps, and prices water at full cost, and publishes the volumes.
- Sets a decibel limit at the property line, names who measures it, and names the fine.
- Puts a floor under community payments, a sunset on abatements, and a bond behind decommissioning.
- Keeps the negotiation in public, with no nondisclosure agreement standing between residents and the terms that will govern their town for 30 years.
- And this is the Toll Road lesson, assigns every major risk to a named party so that when the forecast fails the argument is already settled.
Buchanan’s great insight was that good government is not about finding angels; it is about designing rules under which ordinary, self-interested people produce decent outcomes. A data center is not a gift from the heavens, and it is not a plague. It is a large, thirsty, humming neighbor that wants a 30-year relationship with your town. The question is never whether to welcome it. The question is whether you write the rules of that relationship before it moves in, or spend 30 years discovering that you didn’t.
Barry Keating, Ph.D., an adjunct scholar of the Indiana Policy Review Foundation, is Professor Emeritus of the Mendoza College of Business at the University of Notre Dame.A version of this essay will appear in the winter Indiana Policy Review.
