When I previously argued that America’s AI future depends on abundant, reliable, and affordable energy—not political mandates—I cautioned against allowing data-center policy to become “another exercise in industrial planning driven by political symbolism.” I also noted that there is a profound difference between removing barriers to production and attempting to centrally direct production itself. Pennsylvania’s newly announced Governor’s Responsible Infrastructure Development (“GRID”) Standards illustrate the difference by showing what not to do.
At first glance, Pennsylvania appears to be pursuing a sensible objective. Artificial intelligence will require enormous quantities of electricity. Data centers are becoming critical infrastructure. States that successfully attract AI investment stand to benefit from significant economic growth, infrastructure investment, and high-paying jobs.
New empirical evidence strongly supports that proposition. According to a recent PwC analysis, the U.S. data-center industry supported approximately 5.5 million jobs in 2024 and contributed nearly $927 billion to the national GDP. The scale of that contribution is difficult to overstate. Data centers are no longer a niche component of the digital economy. They are rapidly becoming foundational infrastructure for artificial intelligence, cloud computing, advanced manufacturing, biotechnology, finance, and national security.
At the same time, a recent E3 study undermines one of the most common criticisms leveled against data-center development: the claim that data centers inevitably impose substantial costs on ordinary ratepayers. Examining multiple utility territories, the report found that data centers generally paid more than their incremental cost to serve and, on average, generated net benefits that helped reduce pressure on retail electricity rates. For example, according to the report, in Georgia, data center “growth is expected to place downward pressure on residential bills of $8.50 /month from 2029-2031,” and, in Michigan, “residential costs are expected to lower by 8% due to recently approved data center contracts.” In each case, the reductions are the result of spreading the fixed costs of generation across a larger demand base. In other words, the market is already developing mechanisms to allocate costs through pricing, contracting, and utility tariff structures.
Those findings should matter because they directly challenge the premise that heavy-handed government intervention is necessary to manage AI-related electricity demand. Sadly, Pennsylvania is mired in the government-centric conventional wisdom.
Under Governor Josh Shapiro’s new GRID Standards, data-center developers seeking Commonwealth support must satisfy a lengthy list of state-directed requirements governing energy sourcing, labor practices, community-benefit obligations, environmental standards, reporting requirements, and workforce development initiatives. Most notably, developers must procure increasing percentages of electricity from “clean firm energy” sources, beginning at 10 percent in 2027 and rising to 32 percent by 2035. They also must satisfy various prevailing-wage, compensation, and community-benefit requirements.
The irony is that Pennsylvania correctly identifies the opportunity while simultaneously undermining the conditions most likely to realize it. The central problem is not the goal of encouraging responsible development. It is the assumption that state officials possess sufficient knowledge to dictate the optimal energy mix, labor structure, and investment configuration for an industry evolving at extraordinary speed.
F.A. Hayek famously described this problem in “The Use of Knowledge in Society.” The relevant information necessary for economic coordination is dispersed among millions of individuals and constantly changing. No planner possesses it all. No agency can aggregate it effectively. Markets function precisely because prices communicate information that no single decisionmaker can fully comprehend.
Pennsylvania’s clean-energy quotas embody what Hayek elsewhere described as the “fatal conceit”—the belief that policymakers can successfully direct complex economic systems through centralized planning. Today’s officials are attempting to determine what the energy needs of AI infrastructure will look like in 2035, what technologies will be available, what generation resources will prove most reliable, and what tradeoffs entrepreneurs should make among cost, reliability, and innovation.
They cannot know these things. Nor can they know whether future breakthroughs in nuclear power, advanced natural-gas generation, geothermal systems, battery storage, small modular reactors, or technologies not yet developed will prove superior to the politically preferred mix embedded within today’s regulations.
This is precisely why Austrian economists have long emphasized entrepreneurial discovery. Markets do not merely allocate resources efficiently; they generate knowledge. Innovation often emerges from experimentation that planners neither anticipate nor understand. Regulatory mandates substitute bureaucratic judgments for that discovery process.
The labor and community-benefit provisions raise a different but equally important concern. Public-choice economics, developed most prominently by James Buchanan and Gordon Tullock, rejects the romantic assumption that political actors pursue some abstract conception of the public interest. Politicians, regulators, and organized interest groups respond to incentives just as market participants do.
Viewed through that lens, several aspects of the GRID Standards appear less like neutral public policy and more like familiar political bargaining. Requirements that projects create prevailing-wage construction jobs, provide compensation exceeding statewide averages, and deliver additional community benefits may sound appealing. But they also function as mechanisms for market dislocation. In other words, these requirements distribute concentrated benefits to politically influential constituencies while dispersing costs across developers, investors, and ultimately consumers.
Tullock’s work on rent-seeking is particularly relevant here. When governments condition permits, tax incentives, and regulatory approvals on compliance with politically constructed requirements, firms inevitably devote resources toward satisfying political demands rather than maximizing productive efficiency. Instead of competing to discover the most effective methods of generating electricity, constructing infrastructure, or operating data centers, firms must increasingly compete for regulatory approval.
That process creates incentives for lobbying, coalition-building, and political negotiation rather than innovation. The danger is not merely higher costs. It is slower discovery. Put another way, it’s textbook market dislocation.
One of the striking findings emerging from the AI sector is how quickly markets are adapting. Technology firms are pursuing long-term power-purchase agreements. Utilities are developing large-load tariffs. Investors are financing new generation capacity. Grid operators are exploring innovative approaches to balancing rapidly growing demand. The E3 report suggests that these market mechanisms are already producing workable solutions without requiring comprehensive state planning. When allowed to function, markets will not fail to deliver such results.
Pennsylvania’s policymakers appear unwilling to trust that process. Instead, the Commonwealth has embraced a model that attempts to simultaneously attract AI investment, satisfy environmental constituencies, accommodate organized labor, reassure local communities, and advance broader political objectives. That balancing act may be understandable politically. Governor Shapiro is widely viewed as a potential 2028 presidential candidate, and the incentives facing ambitious elected officials naturally encourage efforts to reconcile competing interest groups.
The problem is that economics is not politics. Markets can accommodate competing preferences because prices reveal tradeoffs. Political systems often obscure those tradeoffs by distributing benefits openly while hiding the costs. Public-choice theory teaches that such arrangements frequently produce results that are less than promised.
Pennsylvania is hardly alone. As states compete to attract AI infrastructure, many will face pressure to attach increasingly elaborate policy conditions to development approvals. Policymakers should resist that temptation.
The lesson of both the PwC and E3 reports is not that the government must become more involved in directing AI infrastructure. It is that the market is already responding. Capital is flowing. Investment is occurring. Infrastructure is being built. New pricing mechanisms are emerging. Entrepreneurs are searching for and finding solutions.
The challenge for policymakers is not to replace those processes with political allocation. It is to remove the barriers that prevent those processes from functioning effectively.
Recent federal policy reflects a markedly different philosophy. The Trump Administration’s AI Action Plan largely focuses on accelerating permitting, expanding generation capacity, preserving reliable baseload resources, and reducing regulatory obstacles to infrastructure development. Simply put, it seeks to increase the supply of energy rather than dictate how that energy must be produced or consumed. That distinction matters as one approach relies on entrepreneurial discovery and market adaptation. The other relies on political allocation. The former recognizes the limits of what policymakers can know, and the latter assumes those limits do not exist.
America needs more electricity. It needs more generation. It needs more transmission. It needs more data centers. Most of all, it needs the institutional humility to recognize that no office, agency, or governor possesses the knowledge necessary to centrally direct the future of AI infrastructure.
Pennsylvania has correctly identified the opportunity. Its mistake is trying to dictate the path.
