The United States currently leads the world in artificial intelligence research, and it is about to squander that lead on a problem as old as the industrial revolution: we are not generating enough electricity to run the machines we are building. The data centers that train frontier AI models are consuming power at a rate that the American grid, in its present condition, simply cannot sustain. This is not a forecast. It is a condition. And continuing to treat it as someone else’s problem to solve, at some later moment, is the clearest sign that a civilization has stopped believing in its own future.
The central fact that should organize American policy right now is this: computational supremacy and energy supremacy are the same race. A nation that produces the most brilliant AI researchers but cannot power their machines will finish second. Finishing second in this contest is not an honorable outcome. It is a strategic catastrophe dressed in polite language about competitiveness and partnerships and managed decline.
There is a tradition in American public life, born out of the great mobilizations of the twentieth century, that understood civilization-scale challenges as invitations rather than burdens. The federal government, in that tradition, did not wait for markets to reach equilibrium before building the interstate highway system, before accelerating the space program, before funding the basic research that eventually gave the world the internet. The argument was never that government should do everything. The argument was that some tasks are too large, too long-term, and too strategically vital to be assembled from quarterly earnings reports alone. That tradition has not been disproven. It has simply been neglected, which is a different thing entirely.
The present diagnosis is not difficult to construct if you are willing to read the numbers honestly. American electricity demand had been essentially flat for two decades before the AI buildout began. Now every serious projection has it climbing sharply. The Department of Energy has acknowledged that data center electricity consumption could double by the end of this decade. Meanwhile, the grid itself is older than the appliances most Americans consider antique. Permitting a new high-voltage transmission line takes, on average, more than a decade. Nuclear power plants, the only carbon-free source capable of delivering the dense, reliable baseload that AI infrastructure requires, have been shutting down faster than they have been opening. The United States has one new large-scale reactor under construction, at Plant Vogtle in Georgia, and the cost overruns and delays at that project have been used, cynically, as an argument against ever building another one, rather than as an argument for building the next one better.
The named enemy here is not China, though China is building nuclear capacity at a pace that should embarrass anyone who calls themselves a serious person in American energy policy. The named enemy is institutional timidity, the refusal to act at the scale the moment requires because acting at scale involves risk, and risk involves the possibility of failure, and failure is politically uncomfortable. So instead the country produces white papers. It convenes summits. It announces frameworks. And the gap between American energy capacity and American computational ambition grows a little wider every quarter.
The prescriptions are not mysterious. First, Congress should pass emergency permitting reform for transmission infrastructure and clean energy generation, with firm statutory deadlines that cannot be stretched by the current maze of jurisdictional review. The permitting process as it exists today was designed for a world in which the grid was adequate. That world is gone. Second, the federal government should treat advanced nuclear, including small modular reactors now moving toward commercial deployment by companies like NuScale and Kairos Power, as a strategic national priority on the same tier as semiconductor fabrication. That means loan guarantees, long-term federal power purchase agreements, and a regulatory pathway at the Nuclear Regulatory Commission that is rigorous without being interminable. Third, the National Science Foundation and the Department of Energy should dramatically expand their joint funding for research into grid-scale storage and next-generation transmission materials, because no single generation technology solves the problem without a smarter, more resilient network to carry its output. Fourth, and this requires more political courage than the others, the administration should be honest with the American public that cheap, abundant, clean electricity is not going to arrive without building things, and that building things means accepting that some of those things will be visible, and loud, and located somewhere specific.
The connection between the AI race and the energy question is not incidental. The countries that will shape the character of artificial general intelligence, whenever it arrives, will be the countries whose researchers had the computational resources to do the foundational work. Training a large frontier model today requires energy that would have powered a small city a generation ago. The next generation of models will require more. If American data centers are throttled by grid constraints while Chinese state-owned utilities pour baseload power into their AI programs without hesitation or permitting delay, the intellectual contest that everyone in Washington claims to care about will have been decided by an infrastructure failure that everyone in Washington saw coming.
The generation that built the great dams and the national laboratories and the early space program did not do so because the task was easy or the politics were simple. They did it because they understood that the size of a nation’s ambitions is a statement about what kind of nation it intends to be. The grid is the new frontier, and a civilization that cannot bring itself to wire it properly has already decided, whatever it tells itself at its own ceremonies, that the future belongs to someone else.
Aurelian Cape is a columnist for The Republic Standard.