Artificial intelligence is often described as a race for chips, algorithms, and talent. Increasingly, however, it is becoming something else: a race for electricity.
The largest AI firms—Meta, Microsoft, Google, Amazon, and OpenAI—are rapidly becoming some of the most consequential energy consumers in the American economy. Advanced AI systems require enormous amounts of continuous computational power, and the data centers supporting them demand vast quantities of reliable electricity. As AI deployment accelerates, the question is no longer whether the United States can build sufficiently advanced models. It is whether the nation can generate enough dependable power to sustain them, while not driving up the cost of electricity for consumers.
To their credit, many hyperscalers appear to recognize the seriousness of the challenge. Meta, for example, has emphasized the need for additional generation capacity, transmission upgrades, substations, and permitting reform while highlighting investments in nuclear energy and other long-term infrastructure solutions as a means for accomplishing this two-pronged objective. The company’s approach suggests an important recognition: AI leadership ultimately depends as much on physical infrastructure as it does on software breakthroughs.
That recognition matters because too much of the current policy conversation surrounding AI and energy remains detached from the realities of electricity production.
For years, policymakers and activists treated energy policy primarily as a climate issue. AI changes the equation. Data centers are not intermittent consumers of electricity. They require stable, continuous-duty power with extremely high reliability. AI training clusters cannot simply power down when the wind slows or cloud cover increases. Continuous uptime and scalable baseload generation are not luxuries for advanced computing infrastructure; they are operational necessities.
This reality should force a broader reconsideration of how America approaches energy production in the age of AI.
Demand Must Be Matched with Supply
The encouraging news is that major technology firms increasingly appear willing to support new supply rather than merely purchase renewable-energy credits or rely on accounting mechanisms that do little to expand actual generation capacity. That shift reflects growing recognition that the nation’s electric grid cannot sustain exponential AI growth without substantial new investment in production and transmission infrastructure.
Still, important risks remain.
Much of the public discourse surrounding AI and energy continues to assume that political preferences regarding “clean” or “green” energy should dictate infrastructure decisions. That assumption risks subordinating engineering realities to ideological priorities. While wind and solar technologies will likely play some role in future generation portfolios, policymakers and market participants should resist efforts to force hyperscalers into narrowly prescribed energy strategies that may prove poorly suited for the unique reliability demands of AI infrastructure.
Markets, not political fashion, are generally better positioned to determine the optimal generation mix.
This is not merely an ideological claim. It reflects a practical reality recognized by economists across multiple traditions. Friedrich Hayek famously warned that centralized planners lack the dispersed knowledge necessary to allocate resources efficiently in dynamic and rapidly changing markets. AI infrastructure presents precisely that kind of environment: technological demands are evolving quickly, regional grid conditions differ dramatically, and future energy requirements remain highly uncertain.
In such environments, decentralized experimentation and entrepreneurial discovery often outperform centralized mandates. Firms responding to price signals, reliability constraints, and competitive pressures are generally more capable of adapting than regulators attempting to forecast the “correct” technological pathway years in advance.
That insight is especially important because AI energy policy is already showing signs of the political dynamics long identified by Public Choice scholars such as James Buchanan and Gordon Tullock. Large subsidy regimes, permitting systems, and industrial-policy programs inevitably create opportunities for rent-seeking behavior in which politically connected industries compete for regulatory advantages rather than for productive superiority. The danger is not simply inefficiency. It is that energy allocation decisions become increasingly shaped by political incentives rather than operational performance and consumer demand.
The reliability issue is particularly significant for AI.
Nuclear power deserves renewed attention precisely because it provides the kind of high-density, continuous-duty generation that advanced computing infrastructure requires. It is encouraging that several major technology firms now openly recognize nuclear energy as an important part of America’s AI future. At the same time, modern natural gas generation remains among the most scalable and reliable energy sources available domestically. Even coal—despite years of political hostility—continues to provide dependable baseload generation in many regions and is substantially cleaner today than in prior decades due to technological improvements.
None of this means renewable technologies should be prohibited or ignored. It means they should compete.
That distinction matters. Political pressure often treats wind and solar deployment as an end in itself, regardless of cost, intermittency, storage limitations, transmission burdens, or lifecycle disposal concerns associated with exhausted batteries, turbine components, and solar panels. Yet these costs are real, even when omitted from public discussion. So too are the geopolitical implications of heavily subsidized global supply chains that have concentrated significant solar panel, wind turbine, battery, and rare-earth mineral production in China.
AI infrastructure policy should not become another exercise in industrial planning driven by political symbolism.
The better approach is simpler: allow firms operating in competitive markets to pursue the generation sources that best satisfy reliability, scalability, and cost constraints. Some regions may favor nuclear. Others may rely more heavily on natural gas, hydroelectric power, geothermal energy, or diversified portfolios that include renewables where economically appropriate. What matters is not whether the resulting mix satisfies ideological preferences. What matters is whether it works.
The Knowledge Problem in Energy Policy
The temptation to centrally manage AI-related energy development is understandable. The scale of projected electricity demand is immense, and policymakers understandably fear shortages, price spikes, and infrastructure bottlenecks. Recent commentary has therefore emphasized large-scale government coordination, industrial policy, and centralized planning mechanisms as solutions to the coming power challenge.
But the history of economic regulation counsels caution.
Complex systems rarely respond well to rigid top-down management. As Austrian economists have long emphasized, markets are not static mechanisms but dynamic discovery processes in which information emerges through decentralized interaction. Prices, investment flows, technological experimentation, and entrepreneurial adaptation collectively transmit information that no centralized planner can fully possess.
Energy systems are particularly complex because they involve constantly shifting interactions among fuel markets, transmission constraints, regional demand patterns, technological innovation, environmental regulations, and consumer behavior. Attempts to centrally optimize such systems often produce unintended consequences, delayed investment, and distorted incentives.
The current permitting regime itself illustrates the problem. It can take years—sometimes more than a decade—to secure approval for major transmission projects, generation facilities, pipelines, or nuclear infrastructure. These delays increase costs, discourage investment, and slow adaptation precisely when rapid expansion is needed most.
Permitting reform may therefore be among the most important AI policies currently available to lawmakers.
Importantly, this is one area where many technology firms and free-market critics of overregulation appear increasingly aligned. Faster approval processes, streamlined transmission development, and reduced regulatory fragmentation would help expand supply without requiring policymakers to dictate specific technological outcomes.
That distinction is critical. There is a profound difference between removing barriers to production and attempting to centrally direct production itself. The first approach enables markets to adapt, driving prices lower. The second risks replacing entrepreneurial discovery with political allocation.
AI’s Energy Future Should Be Built, Not Planned
The United States is entering a new phase of technological competition in which energy abundance may become as strategically important as computational capability itself. Fortunately, the market appears to be responding.
Technology firms are investing heavily in infrastructure, utilities are expanding capacity planning, nuclear power is regaining political legitimacy, and reliability concerns are increasingly impossible to ignore. These developments suggest that the private sector recognizes the scale of the challenge. Policymakers should avoid undermining that adaptation process through overly prescriptive energy mandates or politicized allocation schemes.
The objective should not be to force AI infrastructure into preconceived ideological frameworks. It should be to create conditions under which firms can build reliable, scalable, and affordable energy systems that advanced computing requires. That means reducing regulatory barriers, encouraging infrastructure development, protecting competitive energy markets, and allowing experimentation across multiple generation technologies.
In the end, America’s AI future will depend not only on innovation in Silicon Valley, but also on whether the nation can generate enough dependable, low-cost power to sustain it. The firms building the next generation of artificial intelligence increasingly understand this reality. Policymakers should as well.
