Six days after I published The Cloud Wants to Move into My Back Yard, a piece asking Minneapolis not to approve a proposed North Loop data center on promises alone, the developer walked away.
Legacy Investing withdrew its land-use application for the former Minnesota Star Tribune printing plant and terminated its agreement to purchase the property. The proposed 20-megawatt data center, at least for now, is dead.
I would love to tell you exactly why. I cannot.
Legacy has not publicly explained its decision, and there is no evidence that allows us to say neighborhood opposition killed the project. What we know is that hundreds of people showed up when the proposal was presented publicly in August, many of them strongly opposed. We know the developer had not yet provided the site-specific information people were asking for about electricity, water, noise and other impacts. And we know that, rather than continue through that process, Legacy withdrew its application and its purchase agreement.
In my last piece, I argued that Minneapolis should not say yes until Legacy came back with what I called “the meter, the model and the commitments”: actual resource-use numbers, evidence about impacts and binding promises that could survive a tenant change, expansion or sale.
Legacy never came back with them.
That might sound like the end of the story.
It isn’t.
Minneapolis Kept Going
The data center disappeared. The policy problem did not.
Minneapolis is still developing permanent rules for data centers, and the latest proposal gets surprisingly close to some of the questions I raised in my first piece.
Under the city’s preliminary recommendations, every new or expanded data center would require a conditional-use permit and public hearing. Applicants would have to disclose projected water and electricity use, generator use and plans for controlling sound. They would also have to provide written evidence that they have coordinated with the electric utility.
The city is also proposing something more fundamental: deciding where data centers belong before another developer arrives.
Under the current proposal, new data centers would be limited to downtown zoning districts and placed in buildings that already exist. Downtown server and equipment space would be capped at 175,000 square feet. New purpose-built data center buildings would not be allowed, and data centers established after January 1, 2026 could not be the only principal use on a downtown property.
These rules are not final. The Planning Commission is tentatively scheduled to consider them October 13, followed by City Council consideration in November.
But the direction matters.
Instead of waiting for another proposal and then scrambling to figure out what questions to ask, Minneapolis is beginning to establish the questions in advance. I think that is a much better way to govern infrastructure.
Minnesota Is Asking Who Pays
The state is doing something similar from a different direction.
One of the biggest questions in my previous piece was deceptively simple: if a data center requires new electrical infrastructure, who pays for it?
Minnesota has begun answering that.
State law now requires special treatment of very large electricity customers, and the Minnesota Public Utilities Commission has established a framework intended to keep the costs of serving them from being shifted onto existing households and businesses.
The principle is straightforward: if a data center requires electrical system upgrades and additional infrastructure, the data center should pay the costs attributable to serving it. Utilities must develop special rates for very large customers, and the PUC reviews the agreements governing how those customers are served.
That sounds obvious.
It isn’t how infrastructure costs necessarily work without regulation.
A utility may need to build substations, transmission lines or other equipment based on enormous projected electricity demand. If the customer later scales back, delays construction or disappears entirely, someone can still be left paying for the infrastructure.
That creates another problem regulators are beginning to confront: what happens when the demand itself is speculative?
And Minnesota is no longer dealing with hypothetical projects. The Public Utility Commission (PUC) is reviewing electric-service agreements associated with proposed large data centers, including Google’s planned Hermantown project.
The question has moved beyond whether Minnesota wants data centers.
It is becoming: On what terms?
This Is Becoming a National Question
Minnesota is hardly alone.
In 2026, lawmakers in 16 states introduced or considered legislation that would pause or restrict new data-center development, according to the National Conference of State Legislatures. Some proposals remain active, while others have failed or been vetoed. The stated rationale varies, but many proposals are intended to give governments time to understand effects on electric grids, communities and public resources before additional projects are approved.
Other jurisdictions are trying to regulate rather than pause.
New York has proposed requiring developers of new data centers to make community investments of at least $1 million for every megawatt of utility demand. Massachusetts is requiring large data centers seeking state permits to negotiate community-benefit agreements. Columbus is considering special utility rates for large industrial users along with water conservation, public disclosure and decommissioning requirements.
And this week the issue reached Congress.
On September 16, the U.S. House passed the Ratepayer Protection Act 417 to 3. The legislation directs state utility regulators to address whether data centers and other very large electricity users are paying the costs of the new generation and transmission infrastructure needed to serve them. Critics argue the legislation does not go far enough, but a 417-to-3 vote tells us something about how far this issue has moved into mainstream public policy.
Meanwhile, resistance is growing in communities from Silicon Valley to Texas, New York and the Midwest. The arguments are remarkably similar: electricity, water, noise, pollution, tax subsidies, relatively limited permanent employment and a basic question about whether communities are receiving enough in return for the resources these facilities consume.
A year ago, much of the conversation about AI infrastructure centered on how quickly we could build it.
Now another question is catching up:
Who carries the consequences after it is built?
Approval Is Not the Finish Line
We tend to treat infrastructure decisions as approval decisions. A developer proposes something. Government studies it. Residents comment. Regulators impose conditions. Somebody eventually says yes or no. But approval is only one moment in a system that may operate for decades.
The questions being asked now are largely questions of governance. Who pays for new infrastructure? What limits are placed on expansion? What happens when a tenant changes? What information must be disclosed? Who is responsible for decommissioning, and what obligations survive a sale? These are decisions about the rules under which a data center is allowed to operate and who carries the risk when things do not go according to plan.
Then comes another set of questions that interests me as an evaluator: Did what we were promised actually happen? Did electricity and water consumption stay within projections? Did the developer, rather than existing ratepayers, actually bear the costs attributed to the project? Did promised jobs, tax revenue and community investments materialize? Did noise and emissions remain within agreed limits? Did the benefits reach the people they were supposed to reach? And when actual performance diverged from the projections used to secure approval, did anything happen as a result?
That is where evaluation belongs. A projected water-use number is a prediction, not an outcome. A community-benefit agreement is a promise, not evidence of community benefit. And a ratepayer protection does not, by its existence, tell us whether ratepayers were actually protected. We need to follow these projects long enough to compare what was promised with what happened, understand who benefited and who bore the costs, and identify consequences nobody anticipated when the deal was approved.
If data centers are becoming essential infrastructure, we need both sides of this equation: stronger governance before they are built and meaningful evaluation after they begin operating. The ribbon cutting should not be the moment when public scrutiny ends.
There Are Still Harder Questions
There is a danger, though, in telling this story as communities finally standing up to Big Tech because somebody has to host this infrastructure.
I wrote this before and I think it matters even more now: putting a data center somewhere else does not make its impacts disappear.
If affluent neighborhoods with political influence become very good at stopping data centers, developers will look elsewhere. That could mean rural communities with smaller tax bases, places hungry for investment, communities with weaker regulatory capacity or areas where residents have less time and money to organize.
We could congratulate ourselves for protecting one neighborhood while simply exporting the electricity demand, water consumption, generator emissions and land-use consequences somewhere with less power to resist them.
That is not responsible infrastructure policy.
There is another tension too. Communities can demand so much compensation that community benefits start looking like a price for accepting harm rather than a strategy for preventing it.
New York’s proposed $1 million-per-megawatt community investment is an interesting experiment. A 100-megawatt facility would imply at least $100 million in community investment. That is real money. But money does not make every site appropriate.
A community-benefit agreement cannot manufacture water in a stressed watershed. Tax revenue cannot make an overloaded transmission system reliable. A new park doesn’t cancel out continuous noise next door. Compensation should address impacts that remain after good siting and mitigation, not become permission to ignore them.
And there is a final tension that makes me uncomfortable because I am part of it.
Demand is coming from us.
I use AI heavily. Businesses are embedding it into their operations. Governments are adopting it. People stream more, store more and compute more. AI agents may dramatically increase the amount of computing happening behind a single human request.
We cannot demand infinite digital infrastructure and simultaneously insist that none of its physical infrastructure exist near anyone. But neither should technological demand become a blank check.
Six days after I asked whether one Minneapolis data center could meet that standard, its developer walked away. We still do not know why. The more important development may be what remained after it left.
Minneapolis kept writing rules. Minnesota kept building ratepayer protections. Other states and cities kept experimenting with moratoriums, community-benefit agreements, disclosure requirements, special utility rates and decommissioning plans. Congress took up the question of who should pay for the infrastructure these facilities require.
The conversation is beginning to move beyond whether a data center should receive permission to exist. It is moving toward something harder and more useful:
What did you promise?
What actually happened?
Who benefited?
Who paid?
And what are we going to do if the answers are not the ones we were given?
That is the accountability infrastructure we need to build alongside the data centers themselves.
Anthralytic explores AI, data, evaluation and the systems shaping social impact. We look past both hype and reflexive rejection to ask what the evidence shows, who benefits, who bears the costs and how we build accountability into the systems we create.
