When I first argued in the National Law Review that America should meet artificial intelligence’s energy demands through abundant supply and market adaptation, I warned against turning data-center policy into another exercise in industrial planning. When Pennsylvania later announced its Governor’s Responsible Infrastructure Development (GRID) Standards, I described the program’s clean-energy quotas, labor terms, and community-benefit mandates as the fatal conceit of AI industrial policy. At the time, GRID was presented as the price of obtaining tax incentives, expedited permitting, and coordinated state support. A developer could reject the bargain and proceed under generally applicable law.

That distinction has now largely collapsed. On August 18, Gov. Josh Shapiro signed Executive Order 2026-05, converting GRID from an incentive program into the organizing principle of Pennsylvania’s data-center permitting system. The fatal conceit now comes with a 33-page consent order, stipulated civil penalties, and waivers of legal challenge.

The timing is difficult to miss. Public opposition to data centers has surged as the fall elections approach. The Wall Street Journal recently identified governors of both parties who once courted data centers and now campaign against them. In Pennsylvania, Shapiro has moved from celebrating Amazon’s $20 billion investment to denouncing “greedy,” “predatory” developers and promising the “strictest guardrails in the nation.” His signing remarks repeatedly invoked the “full weight” of executive authority and told noncompliant developers to go elsewhere.

Pennsylvania has not enacted a formal moratorium. Its new system is subtler and, for that reason, may prove more consequential. The executive order creates an administrative permission structure in which the Commonwealth controls the order, speed, and cost of nearly every development workstream. A project may not be prohibited in name, but government can make the disfavored path slow, serial, uncertain, and commercially impractical.

Consider the two permitting lanes established for projects with peak demand greater than 25 megawatts. A developer that signs a project-specific consent order implementing GRID may receive “rolling” review from the Department of Environmental Protection (DEP). In practical terms, DEP can review separate permit applications as they arrive and, after local approvals are documented, issue qualifying permits without waiting for the entire package to be completed. Thus, an application for one component of a project need not sit untouched merely because a different authorization remains under development. A developer that declines GRID receives markedly worse treatment. DEP may not even begin reviewing its applications until the project secures local approvals and any required water-withdrawal or wastewater authorization. The agency may not issue permits incrementally, must wait until every required application has been received and reviewed, and must exclude the project from the Commonwealth’s PAyback and Permit Decision Guarantee programs. All data centers, including GRID participants, are removed from Fast Track permitting, and the sales-and-use-tax exemption is reserved for GRID-compliant projects.

This matters because the governor’s rhetoric and the legal text do not quite match. Shapiro declared that GRID is “not voluntary” and that a developer refusing its requirements will have “no pathway” to the approvals needed to build. Yet the order expressly addresses applicants that have not signed a GRID consent order. As Polsinelli’s careful analysis explains, GRID is not stated as a legal prerequisite to development, but refusing it may carry substantial schedule consequences. In other words, the Commonwealth preserves a formal choice while loading the alternatives. The issue is not a literal ban. It is coercion through administrative sequencing.

The template consent order makes the scope of that leverage clear. A developer must “build, bring, or buy” incremental capacity sufficient to satisfy peak demand and incremental energy sufficient to cover annual consumption. It must obtain an increasing share of its electricity from politically defined “clean firm” sources located in Pennsylvania—10% beginning in 2027, 14.5% in 2030, and 32% in 2035—or make an alternative-compliance payment. Buildings over 100,000 square feet must be solar-ready.

The order reaches well beyond electricity and environmental performance. It requires at least 200 prevailing-wage construction jobs, 50 permanent jobs paying at least 125% of the statewide average wage, at least $1.5 million in annual compensation after the fourth year of operation, and at least $250 million in cumulative new investment. A developer must offer municipal and county governments a community-benefit agreement that can include financial contributions and “other items” those governments determine. The resulting obligations may be folded into DEP permits, and the developer agrees not to challenge their inclusion.

Violations can trigger automatic penalties of $25,000 per day. Violations of the energy obligations can produce penalties between $25,000 and $100,000 per megawatt of rated capacity per day. The developer waives the right to challenge the consent order’s content or validity and waives administrative appeals from decisions made under it. DEP may approve or disapprove required submissions, demand revisions on a schedule it specifies, and make the approved documents enforceable parts of the agreement.

These provisions raise a basic legal and institutional question: What is the relationship between the environmental statutes cited as authority for the consent order and the state’s commands concerning electricity procurement, wages, aggregate compensation, investment levels, and community bargaining? The executive order repeatedly adds the phrase “to the extent permitted by law,” but a savings clause does not answer the question. Nor does nominal consent eliminate it when the alternative is delayed review, lost timing protections, exclusion from Fast Track, and forfeiture of a valuable tax exemption.

Some parts of the order are defensible. State officials should not sign broad nondisclosure agreements that prevent them from informing the public about matters they regulate. A public map of proposed projects can improve transparency. Utilities should assign interconnection and dedicated-infrastructure costs to the customers that cause them. Large loads should post adequate security, make minimum payments when utilities build capacity on their behalf, and bear exit costs if a project disappears. Local governments may enforce neutral rules governing noise, traffic, setbacks, lighting, water, and emergency services. These are ordinary applications of property rights, cost causation, and the rule of law.

But legitimate safeguards do not justify turning DEP into a super-agency for energy procurement, labor policy, economic development, and municipal bargaining. The central Austrian insight is not that every project is good or that every corporate promise should be believed. It is that knowledge about the right project, technology, location, contract, and timing is dispersed and often does not exist until entrepreneurs test alternatives.

A data-center project requires simultaneous coordination among site control, zoning, water, environmental permits, utility service, generation procurement, equipment orders, customer commitments, and financing. The order instead forces key workstreams into sequence, particularly for developers that reject GRID. But one approval often depends on confidence that another can be obtained. A developer may be reluctant to make binding power commitments without permit visibility; a generator may be reluctant to finance new capacity without a credible customer load; and a customer may be unwilling to commit without a reliable construction schedule. By delaying information from one process until another is complete, the Commonwealth does not merely slow development. It prevents market participants from discovering whether the pieces can fit.

Capital markets are already performing a screening function. The New York Times reports that the largest AI companies have issued more than $200 billion in debt this year and that borrowing spreads have risen as investors absorb the supply. That is not evidence that capital is free or that every speculative project will be financed. It is evidence that prices are adjusting. Higher borrowing costs force developers and hyperscalers to distinguish valuable projects from marginal ones, accept tighter terms, postpone construction, or abandon plans that cannot cover their costs. A political permission system substitutes administrative favor for that continuing market discipline.

The market is also changing the underlying technology. Eaton’s Pittsburgh-based electrical division and Trane Technologies recently announced an integrated power-and-cooling design that the companies say can reduce internal copper use by as much as 80%, improve energy efficiency by 15%, and cut installation costs by 30% compared with traditional lower-voltage designs. Those are company estimates, not guaranteed outcomes. Still, they illustrate entrepreneurial discovery: scarcity in power, cooling capacity, copper, and time creates incentives to redesign the system. No Pennsylvania plan specified that response.

The benefits of that discovery do not stop at the data-center fence line. GE Vernova recently announced nearly $166 million in additional investment in its Pittsburgh-area operations, including an expanded Speers facility and a new Rostraver plant focused on high-voltage equipment, with hundreds of jobs expected over the next several years. GE Vernova serves a broad energy market, so the investment cannot be attributed to data centers alone. But it illustrates the downstream industrial opportunity: rising demand for electricity and grid equipment can support Pennsylvania manufacturing, skilled employment, and additional productive capacity. Policies that suppress the load also risk suppressing the supplier investments that would help the Commonwealth meet it.

Pennsylvania’s treatment of brownfields reveals the order’s internal tension. The order directs DEP to identify ways to expedite clean generation and energy-storage projects on brownfields and previously developed sites. Yet it separately removes every data center from Fast Track, including two proposed data centers on brownfields: Project Hummingbird on a former coal site in Greene County and the Sharon Data Center on a former Westinghouse site in Mercer County. Hummingbird would pair the data center with onsite generation and battery storage; Sharon reports that it has secured electrical capacity. The order, therefore, may accelerate a qualifying generation or storage component while slowing the data-center project whose demand and capital could make that component economic. Pennsylvania is not regulating unrelated industries here. It is favoring one side of the same transaction while burdening the other.

The unseen costs extend to workers and suppliers. In a recent New York Times guest essay, a representative of IBEW Local 26 reported that annual hours worked by members in the Washington, D.C., region doubled from about 14 million a decade ago to 28 million in 2025, with 600 new apprentices added last year. The author attributes much of that growth to data-center construction and maintenance. Those figures come from a labor advocate and do not prove that every facility produces comparable benefits. They do show why permanent server-room headcount is an incomplete measure. Data centers create demand for electricians, generation equipment, cooling systems, fiber, engineering, maintenance, construction, and tax-supported public services.

Public Choice economics explains why those benefits receive less political weight than opposition at a zoning hearing. The noise, traffic, land-use change, or feared utility increase is visible and concentrated. The supplier that never expands, apprentice who is never hired, generation project that never becomes financeable, and innovation that does not occur are dispersed and unseen. Organized opponents know which project they want stopped. Future beneficiaries may not know that the opportunity existed.

Public Choice analysis does not require mind-reading. It does require calling the maneuver by its proper name. Recent reporting describes data centers as a potent bipartisan campaign issue, and Pennsylvania polling shows overwhelming local opposition. Shapiro’s rhetoric hardened as the issue became politically valuable. When the state Senate declined to enact GRID, his administration used executive control over permit timing, agency programs, and tax administration to approximate the policy the legislature had not enacted. That is not merely technocratic caution. It is an executive end run: using administrative choke points to impose substantive conditions that could not command legislative approval. Public Choice theory predicts precisely this migration from general rules to discretionary leverage when visible political rewards are immediate and the economic costs arrive later.

The concern is not abstract, and Shapiro is not a detached observer of Pennsylvania’s recent conflicts over executive power. As attorney general, his office defended Gov. Tom Wolf’s pandemic business-closure order, arguing that the governor possessed the emergency authority being challenged. After the Wolf administration’s prolonged pandemic emergency, Pennsylvanians did not merely wait for a change in office. They amended Article IV of the state Constitution to place disaster declarations under a 21-day limit unless the General Assembly affirmatively extends them and to prohibit the governor from reissuing substantially similar declarations without legislative approval. A constitutional amendment is an extraordinary institutional rebuke to executive overreach.

Against that history, the Shapiro administration’s response to the Pennsylvania Supreme Court’s separate limitation on agency authority is especially revealing. Pennsylvania’s Department of Health is proposing a major revision to its communicable-disease regulations. The official regulatory analysis expressly identifies the court’s 2021 decision invalidating the department’s school-mask order as the reason to broaden the purposes supporting “any other disease control measure” to include prevention, containment, or mitigation. The proposal also rewrites investigative authority to expressly encompass apartments, buildings, schools, colleges, universities, and other locations, while broadly forbidding obstruction or interference with investigations.

Public-health regulation presents different facts and must be judged under different law. But the method is familiar: when a court or legislature confines or refuses to grant executive power, treat the constraint as a drafting problem and reconstruct the discretion through regulation. Paired with the data-center order, the pattern supports a harder conclusion than mere concern about administrative flexibility. It is an authoritarian governing method—concentrating policymaking in executive agencies, attaching sweeping conditions to permission the state controls, and confronting resistance with what Shapiro himself calls the “full weight” of executive authority. Austrian economics warns that officials cannot possess the dispersed knowledge such power presumes. Public Choice warns that they will exercise it under political incentives, not as disinterested guardians of a knowable public interest.

Pennsylvania has a better option. Rules for large loads should turn on measurable characteristics rather than the customer’s industry label, political unpopularity, or willingness to enroll in the governor’s preferred program. Utilities and regulators can require any sufficiently large customer to demonstrate site control, post deposits, meet milestones, substantiate demand forecasts, fund dedicated upgrades, and secure obligations against cancellation or insolvency. Tariffs can price capacity, transmission, distribution, backup service, and curtailment rights. Projects that can cover those costs proceed; projects that cannot reveal themselves as uneconomic.

Environmental and land-use rules should work the same way. Set measurable limits for emissions, noise, withdrawals, wastewater, traffic, and setbacks, then apply them to comparable facilities. Require truthful disclosure and predictable community participation, but decide applications within defined timelines. If Pennsylvania believes its data-center tax exemption cannot be justified under a neutral tax system, it should repeal the preference rather than retain it as leverage for an expanding industrial-policy program.

Most important, government should establish general, ex ante rules governing measurable harms and cost responsibility—not specify either the commercial outcome or the means of production. Requiring every sufficiently large customer to pay the incremental grid costs it causes is not a state-selected outcome; it is a neutral rule of cost causation. Once those rules are set, entrepreneurs, utilities, investors, workers, and consumers should determine the generation mix, wage structure, investment scale, and contractual arrangements. Market outcomes emerge from that process of discovery. They are not administrative targets.

The warning extends far beyond Pennsylvania. In July, Gov. Kathy Hochul imposed a one-year moratorium in New York, withholding discretionary environmental permits not already deemed complete for hyperscale data centers while state agencies develop comprehensive standards and a community-investment framework. In August, Gov. Greg Abbott directed Texas regulators to audit every data-center project advancing through ERCOT’s interconnection process before any project moves forward and to deny grid access to projects that fail the audit. ERCOT’s queue plainly requires verification: the governor reports more than 474 gigawatts of connection requests, roughly 90% associated with data centers. But the extraordinary size of a speculative queue justifies deposits, milestones, cost causation, and project-specific review. It does not justify freezing every project while political officials devise another layer of discretionary control.

The three states have chosen different legal mechanisms, but they expose the same political temptation. New York calls it a moratorium. Texas calls it an audit. Pennsylvania calls it a consent order. Democratic and Republican governors alike receive an immediate electoral reward for announcing that they have stopped a visible and unpopular project category, while the costs of lost generation, construction, innovation, and tax base emerge later and fall on people who may never know what was lost. That bipartisan convergence is Public Choice in action, not evidence that central planning has suddenly become wise.

In my recent Truth on the Market essay, I argued that a moratorium builds no generation, repairs no transmission line, and shortens no interconnection queue. New York’s literal pause, Texas’s audit freeze, and Pennsylvania’s permission system all suffer from that defect. They control the applicant without solving the scarcity. They may suppress visible projects while leaving electricity supply, permitting delays, and national demand for computing untouched.

Pennsylvania once had the right diagnosis: AI requires more power, infrastructure, and investment. Its mistake was first to prescribe the path and now to condition government permission on following it. The Commonwealth can be a laboratory for entrepreneurial discovery or a laboratory for administrative control. It cannot be both. Other states face the same choice. They should adopt neutral rules that make large customers bear measurable costs and harms, then allow prices, contracts, and entrepreneurship to discover which projects work.

The fatal conceit does not stop at Pennsylvania’s border. Whether it arrives as a moratorium, an audit, or a consent order, it conditions, delays, penalizes, and commands.

Disclaimer: The opinions and views expressed in this article are those of the author and not necessarily those of The National Law Review.