AI Data Centers Outgrow Local Judgment
The first sign that something fundamental had shifted was not a headline about artificial intelligence. It was a quiet admission from a governor who had spent years courting the same industry she now wanted to slow down. Kathy Hochul of New York had once insisted that data center decisions belonged to local municipalities, that the state should not interfere with land use choices made by communities closest to the projects. In July 2025, she signed an executive order imposing a one-year moratorium on hyperscale AI data centers, those requiring 50 megawatts or more to operate. [1] The reversal was not a change of heart. It was a change of calculation.
What had changed was not the technology. What had changed was the realization that local communities were not equipped to negotiate with companies whose annual revenues exceed the gross domestic products of entire nations. Hochul said as much when she explained her reversal: local officials lacked the negotiating ability, the clout, the wherewithal to secure meaningful benefits or to ensure that these corporations either brought their own power or paid a premium into the grid. In other words, the judgment required to evaluate a data center proposal had become too complex for the people who were supposed to exercise it. The human skill of weighing costs against benefits, of understanding what a community gains and what it loses, had been rendered insufficient by the sheer scale of what was being proposed.
This is the pattern that repeats across the country, state by state, as elected officials confront the same discovery. The judgment that once seemed straightforward — should we welcome this investment? — has become something that no single human perspective can adequately handle. Josh Shapiro of Pennsylvania learned this when he traveled his commonwealth and heard from residents who felt bullied by developers. He had once celebrated Amazon’s $20 billion data center announcement as the largest private sector investment in Pennsylvania history, framing it as an opportunity for workers and revenue for communities. [6] By August 2025, he was announcing what he called the strictest guardrails in the nation, requiring developers to secure local approval before receiving state permits. [2] When asked what prompted his 180-degree turn, he did not cite new data or expert analysis. He said he listened. He said he got the vibes of a community.
The phrase is telling. Vibes are not analysis. Vibes are the accumulated impressions that human beings form when they are present in a place, when they hear the concerns of people who will live with the consequences of decisions made far away. Shapiro discovered that the formal processes of evaluation — the permitting systems, the environmental reviews, the economic impact assessments — had missed something that ordinary human perception could catch. The developers had followed the rules. The rules were inadequate to the reality they were meant to govern.
What makes this moment distinct is not that politicians are changing their minds. Politicians change their minds constantly, often for reasons that have nothing to do with principle. What is distinct is the mechanism of change. These officials are not responding to new information about data center economics or grid capacity or water consumption, though such information exists in abundance. They are responding to the failure of their own judgment to account for what their constituents were experiencing. The expertise they thought they had — the ability to evaluate a major economic development proposal — turned out to be insufficient for the task.
Consider Katie Hobbs of Arizona, who pushed state lawmakers to eliminate a tax incentive for data center development that she called a thirty-eight million dollar corporate handout. Hobbs was open about her own history: she voted to create the incentive more than a decade ago when she served in the state legislature. At the time, the incentive made sense. Arizona was not a national leader in data centers. The tax break was a strategic investment in building an industry that did not yet exist in the state. But Hobbs now argues that the industry has matured, that Arizona has become a national leader in the sector, and that taxpayers should not continue subsidizing an industry that can stand on its own.
The logic is clear enough. But the deeper question is why Hobbs’s original judgment — that the incentive was worth creating — did not include a mechanism for revisiting that judgment once its purpose was achieved. The tax break was designed to attract an industry. It was not designed to recognize when the industry no longer needed attracting. The skill of evaluating a policy’s ongoing relevance, of distinguishing between an investment and a subsidy, between a temporary incentive and a permanent entitlement, proved to be beyond the capacity of the original decision. The judgment was made. The judgment was not revisited. And now the judgment has to be unmade, painfully, in public, by the same people who made it.
JB Pritzker of Illinois offers another variation on the same theme. In June 2025, he moved to suspend new tax incentives for data centers, calling on the state legislature to advance reforms that would ensure developers pay their fair share for power use, minimize water use, and be more transparent. [3] This from a governor who, in 2023, celebrated Meta’s $1 billion data center in DeKalb as bringing investment and vitality to a newly thriving community. [7] His spokesperson explained the reversal with remarkable candor: responsible governance means reassessing incentives when circumstances change. The conversation around data centers has changed with the advent of AI hyperscalers, the spokesperson said. The industry that Pritzker once welcomed is not the industry that now wants to build in Illinois.
This is the crux of the matter. The judgment that these governors exercised when they welcomed data centers was based on an understanding of what data centers were. That understanding is now obsolete. The AI data centers being proposed today are not the server farms of a decade ago. They require enormous amounts of power — the 50 megawatt threshold that Hochul used in her moratorium is roughly the amount of electricity needed to power 40,000 homes. They consume vast quantities of water for cooling. They strain grids that were not designed for their demands. And they are being proposed at a scale that transforms the communities that host them, not merely adding jobs or tax revenue but fundamentally altering the character of the place.
Wes Moore of Maryland has taken to describing Virginia, home to the famous Data Center Alley, as a cautionary tale. The way they did it there cannot and will not happen in Maryland, Moore said in August 2025. [4] This is a remarkable statement from a governor who, in 2024, signed legislation removing regulatory barriers to technology infrastructure, praising the bill for supercharging the data center industry in Maryland. Moore’s trajectory mirrors that of his colleagues: enthusiasm followed by exposure to consequences followed by retrenchment. But his invocation of Virginia suggests something more than political expediency. It suggests that the human capacity to learn from the experience of others is still functioning, even as the technological capacity to outpace that learning accelerates.
Greg Abbott of Texas represents the Republican version of this phenomenon. In August 2025, he announced an effective pause on data centers seeking to connect to the state’s power grid until the Public Utility Commission and the Electric Reliability Council of Texas conduct a comprehensive audit. [5] The order cited the need to ensure that ERCOT can keep up with future demand. The numbers are stark: 90 percent of new power requests in Texas are for data centers. [5] Abbott, who previously bragged about his state’s ability to attract big projects and celebrated Google’s planned $40 billion investment, now finds himself in the position of slowing the very development he once championed. [8]
What unites these governors across party lines and regional differences is not a shared ideology. It is a shared discovery about the limits of their own judgment. Each of them believed, with good reason, that welcoming data centers was the right call for their states. Each of them acted on that belief, offering incentives, streamlining permitting, celebrating announcements. And each of them has now discovered that the judgment they made was based on incomplete information, not because they failed to do their homework but because the homework itself was inadequate to the task.
The problem is not that these officials were foolish or corrupt or captured by industry interests, though such failures certainly exist in American politics. The problem is more fundamental. The problem is that the scale and speed of AI development have outpaced the human capacity to evaluate its consequences. A governor in 2020 could reasonably assess a data center proposal by looking at jobs, tax revenue, and community impact. A governor in 2025 cannot, because the proposal before them is not for a data center in the traditional sense. It is for a piece of infrastructure that will consume as much power as a small city, that will strain the grid, that will affect electricity rates for every resident, that will consume water in drought-prone regions, and that will do all of this for a period of decades, long after the current governor has left office.
This is where the question of superfluity becomes urgent. The judgment that these governors are struggling to exercise — the evaluation of whether a major technological investment serves the public interest — is precisely the kind of judgment that AI is increasingly being asked to make. Algorithmic systems are already being deployed to assess everything from loan applications to parole decisions to hiring practices. The promise of these systems is that they can process more information than any human, that they can identify patterns that human analysts miss, that they can make decisions more consistently and more fairly than the flawed, biased, inconsistent humans they replace.
But the experience of these governors suggests a different possibility. The problem with their original judgments was not a lack of information. The problem was that the information they had was systematically incomplete, that the frameworks they used to evaluate data centers did not account for the full range of consequences that would follow. An algorithm trained on the same frameworks would make the same mistakes, only faster and with more confidence. The bias is not in the human. The bias is in the framework. And no amount of computational power can correct for a framework that asks the wrong questions.
This is the deeper lesson of the data center backlash. The governors who welcomed these projects were not stupid. They were using the best available frameworks for evaluating economic development. Those frameworks were developed in an era when the projects being evaluated were smaller, simpler, more contained. The frameworks have not caught up to the reality of what is being proposed. And the people who rely on those frameworks, whether they are governors or algorithms, will make the same errors until the frameworks themselves are revised.
The irony is that AI could help with this revision. An algorithmic system could, in principle, model the long-term consequences of a data center proposal with far more accuracy than any human analyst. It could account for grid capacity, water availability, climate projections, economic multipliers, community impacts, and a hundred other variables that no single human could hold in their head simultaneously. The technology exists to make better judgments than the governors are currently making.
But the technology is not being used that way. The technology is being used to build the data centers, not to evaluate them. The same AI systems that require enormous computational infrastructure are being deployed to optimize supply chains, manage energy grids, and predict consumer behavior. The capacity for sophisticated modeling that could inform better public decisions is instead being directed toward maximizing corporate profits and accelerating the very development that communities are now pushing back against.
The contradiction is not lost on the governors who are struggling with these decisions. They are being asked to evaluate proposals for infrastructure that will house the very systems that could help them evaluate those proposals. The tool that could improve their judgment is the reason their judgment is needed in the first place. The technology that could make them better decision-makers is the technology that has made their decisions more difficult.
This is the sense in which AI makes human judgment superfluous, but not in the way that the technology’s enthusiasts imagine. The fear that AI will replace human decision-makers assumes that AI will be better at making decisions than humans are. In some narrow domains, this is already true. An algorithm can process more loan applications in an hour than a human loan officer can in a week, and it can do so with fewer errors and less bias. An algorithm can review more medical images than a radiologist can, and it can flag anomalies that a tired human eye might miss. In these cases, the human judgment being replaced is genuinely inferior to the algorithmic judgment that replaces it.
But the data center backlash reveals a different dynamic. Here, the human judgment being exercised is not inferior to what an algorithm could do. It is inferior to what an algorithm could do if the algorithm were directed at the right problem. The governors are not failing because they are worse than machines at processing information. They are failing because the information they need does not exist, because the consequences of their decisions are too distant and too diffuse to be fully grasped at the moment of decision. An algorithm could help them see those consequences. But the algorithm is not being built for that purpose. It is being built for the purpose of making data centers more efficient, which makes the data centers more attractive, which makes the decisions harder, which makes the need for better judgment more acute.
The governors are caught in a loop that they did not create and cannot escape. They are being asked to make judgments about a technology that is advancing faster than their ability to understand it. They are being asked to weigh costs and benefits that are not fully known and may not be knowable in advance. They are being asked to act as stewards for communities that will live with the consequences of their decisions for decades, even as the pace of technological change makes those consequences increasingly unpredictable.
The response of the governors has been to slow down. Hochul imposed a moratorium. Shapiro established guardrails. Hobbs eliminated a tax break. Pritzker suspended incentives. Abbott paused grid connections. Moore set conditions for his support. These are not solutions. They are pauses, moments of reflection in which the governors are acknowledging that they do not know enough to make the decisions that are being demanded of them. The pauses are an admission of ignorance, a recognition that the judgment required exceeds the judgment available.
This is the real story of AI and judgment. The technology does not simply replace human decision-makers with better algorithmic ones. It changes the nature of the decisions that need to be made. It creates situations in which the consequences of decisions are more far-reaching, more interconnected, and less predictable than they have ever been. It demands judgments that no human is equipped to make and no algorithm is currently designed to support. And it does all of this at a pace that leaves little time for reflection, for learning, for the slow accumulation of wisdom that has traditionally guided human judgment.
The governors who are now reversing their positions on data centers are not hypocrites. They are learners. They are people who made the best judgments they could with the information they had, who then encountered new information that changed their understanding, and who had the courage to admit that they had been wrong. This is not a failure of judgment. It is the essence of judgment. The capacity to revise one’s conclusions in the face of new evidence is what distinguishes genuine wisdom from mere opinion.
But this capacity is under threat. The pace of technological change is accelerating to the point where the learning cycle can no longer keep up. A governor who welcomes a data center in 2024 discovers by 2025 that the consequences are not what they expected. But by then, the decision has been made, the infrastructure is being built, and the community is committed to a future that it did not fully choose. The pause that Hochul imposed in New York, the guardrails that Shapiro established in Pennsylvania, the audits that Abbott ordered in Texas — these are attempts to create space for learning in a system that provides no space at all.
The question that remains is whether that space will be used well. The governors are pausing, but they are not solving. They are acknowledging that their judgment was inadequate, but they are not articulating what better judgment would look like. They are slowing the development of data centers, but they are not developing the frameworks that would allow them to evaluate data centers wisely. They are, in effect, admitting that the human skill of judgment has been rendered insufficient by the scale of what they are being asked to judge, without yet having a replacement for that skill.
The technology that could provide that replacement exists. AI systems can model complex systems, simulate long-term consequences, and identify trade-offs that human analysts might miss. The capacity for better judgment is available. But it is not being directed at the problem of public decision-making. It is being directed at the problem of corporate efficiency. The same technology that could help governors make better decisions about data centers is being used to make data centers more profitable, which makes the governors’ decisions harder, which increases the need for the technology that is not being deployed.
This is the contradiction that remains even as the technology improves. The better AI becomes at modeling complex systems, the more it enables the construction of infrastructure that requires complex judgments. The more sophisticated the algorithms become, the more they accelerate the pace of change that outruns human understanding. The more capable the technology becomes, the more it demonstrates the inadequacy of the humans who are supposed to govern it.
The governors who are pausing data center development are not Luddites. They are not opposed to technology or progress or economic development. They are people who have discovered, through direct experience, that the judgment required to govern AI infrastructure exceeds the judgment available to them. They are responding to that discovery in the only way that responsible leaders can: by slowing down, by asking questions, by demanding more information before making commitments that will bind their communities for decades.
But slowing down is not the same as catching up. The governors are buying time, but they are not using that time to build the capacity for better judgment. They are not investing in the modeling and simulation tools that would allow them to evaluate data center proposals with greater accuracy. They are not developing the frameworks that would distinguish between beneficial and harmful development. They are not creating the institutions that would allow for ongoing assessment rather than one-time decisions.
The skill that is becoming superfluous is not judgment itself. It is the particular kind of judgment that has traditionally guided public decision-making: the ability to weigh competing interests, to anticipate consequences, to learn from experience, and to revise conclusions in the face of new evidence. This kind of judgment is not obsolete. It is more necessary than ever. But it is no longer sufficient. The scale and speed of technological change have outpaced the human capacity to exercise this kind of judgment effectively.
The governors who are reversing their positions on data centers are demonstrating that human judgment still works. They are showing that people can learn, that they can change their minds, that they can respond to new information and new experiences. This is cause for hope. But it is also cause for concern. The learning is happening too slowly, and the consequences of getting it wrong are too large. The governors are learning, but they are learning at the expense of communities that will live with the results of their earlier judgments for decades to come.
The path forward is not to abandon human judgment in favor of algorithmic decision-making. The path forward is to augment human judgment with the tools that AI can provide, to build systems that help humans see the consequences of their decisions before those consequences become irreversible. The technology exists. The question is whether it will be directed toward that purpose, or whether it will continue to be directed toward the acceleration of the very development that requires better judgment in the first place.
The governors who have paused data center development are not the story. They are the symptom. The story is the discovery that human judgment, as it has traditionally been exercised, is no longer equal to the task of governing the technologies that humans have created. The story is the recognition that the skills that brought us to this point — the ability to evaluate, to decide, to commit — are not sufficient for what comes next. And the story is the open question of what will replace those skills, or whether they can be augmented fast enough to keep pace with what they have unleashed.
The answer to that question will determine not just the fate of data centers or the future of AI, but the nature of human governance in an age of technological acceleration. The governors are pausing because they do not know what to do. The technology that could help them know is available but not directed at their problem. And the communities that will live with the consequences of their decisions are waiting, as communities have always waited, for the people in charge to figure out what they are doing.
The judgment that no longer needs a judge is not the judgment that has been automated. It is the judgment that has been outscaled, rendered inadequate by the very technologies it was supposed to govern. The governors reversing their positions are not being replaced by algorithms. They are confronting the limits of their own capacity to understand what they have unleashed. That confrontation, uncomfortable as it is, may be the most important human judgment being exercised anywhere in the world today.
Sources
1. Kathy Hochul
2. Josh Shapiro
3. JB Pritzker
4. Wes Moore
5. Greg Abbott
6. Amazon
7. Meta
8. Google