The rapid rise of artificial intelligence has both positive and negative consequences, many of which have yet to be fully understood. But the explosive growth in its use is indisputable. One critical domain of that growth is national defense. We are in an accelerating race with China, and the competition includes the three principal components of AI: algorithms, the data used to train models, and the computing infrastructure.
Algorithms are the instructions or rules that govern how the data are organized, analyzed, and interpreted. The data involved in defense are massive, including images, signal intelligence, and real-time tracking of movements and assets obtained from sensors, satellites, networks, and more. All of this must be ingested and processed to establish millions of possible scenarios, which are then simulated billions of times to produce distilled guidance for decision makers. And it’s all constantly changing.
The computing power (chips) and electric power are the most critical parts of the infrastructure required. The power requirements are massive, and this is a crucial element of the U.S.’s competition with China. The PRC already generates nearly 2 1/2 times our domestic power output, and its coal generation alone exceeds our total.
Numerous analyses, including one by the International Energy Agency in April 2025, have projected that electricity demand from AI data centers worldwide will more than double by 2030. That may be conservative. In the United States, AI is on track to consume more electricity in 2030 than power for manufacturing all energy-intensive goods combined, including aluminum, steel, cement, and chemicals.
Generating the necessary power is essential to the exploitation of AI for national security. The effectiveness of our digital arsenal is determined by the number of scenarios that can be processed rapidly and simultaneously to facilitate military planning and operations, as well as to hack, track, or evaluate opposing AI.
We and others have argued that nuclear energy is the cleanest, most reliable, and most continuous power source. This is hardly news: The U.S. Navy boasts some 160 operating nuclear-powered ships. The first, the USS Nautilus, was commissioned in 1954. As we concluded in other analyses, only nuclear energy in the form of mass-produced small modular and microreactors will be able to provide sufficient and affordable power for the nation’s needs. But in the immediate future, we will need fossil fuels to boost capacity rapidly.
Our analysis has not even touched upon how AI and its electricity appetite will affect many other aspects of our economy and everyday lives. But therein lies the “deadly delusion” to which we refer: the belief that near-term decarbonization of power generation can happen simultaneously with building adequate military capability and satisfying civilian applications.
We have previously addressed some of the relevant considerations in this dilemma. We have shown with hard data that the cost of repurposing solar and wind power from intermittent generation into continuously available power, which is needed for AI, makes them untenable for that purpose. Their direct costs are not competitive with fossil fuels or nuclear without massive subsidies, never mind second-order costs.
We have also shown how electric vehicles have minimal impacts on emissions. Yet we waste subsidies and resources on sprawling charging and local grid infrastructure to support increased EV penetration instead of redirecting them to military AI. Meanwhile, when small-scale nuclear power generation or new natural gas plants and AI data centers can be collocated, we spend less on power transmission infrastructure.
AI behemoths, including Google, Microsoft, Amazon, Oracle, Meta, and Palantir, are already embracing nuclear energy, especially small modular reactors and even fusion prototypes, to meet the burgeoning electricity demands of their AI data centers. In a few cases, they are reopening large, decommissioned, conventional nuclear plants.
Moreover, many people seem oblivious to the fact that Western nations are responsible for only about 25% (and a decreasing fraction) of worldwide emissions. We are doing our part on climate change and cannot allow climate obsessions to put us at a military disadvantage versus adversaries who emit greenhouse gases with impunity. The reality is that AI capability and capacity will likely determine military security and dominance for the near future.
SALESFORCE CEO IGNITES FIRESTORM AFTER LAYOFFS AND PRAISE FOR AI
Bottom line: We risk crippling our military and economy with wasted overinvestment in intermittent energy sources and EVs, by failing to invest in extensive cloud computing and power generation as needed, and by limiting the production of fossil fuels to meet immediate needs. Unrealistic policies can make the country less powerful, in both senses of the word, and, therefore, more vulnerable. We are far less threatened by rogue AI than by the superior cyber expertise and AI capability of our adversaries.
The demands of AI have radically changed the landscape underlying energy policy, and we can ill afford to continue the diversion of money into ineffective or symbolic homage to the climate gods. That is the deadly delusion.
Andrew I. Fillat spent his career in technology venture capital and information technology companies. He is also the co-inventor of relational databases. Henry I. Miller, a physician and molecular biologist, is the Glenn Swogger Distinguished Scholar at the Science Literacy Project. They were undergraduates together at M.I.T.