AI is turning electricity into ammunition

AI is turning electricity into ammunition. The American military's appetite for fuel is famous enough to travel as a statistic without its footnotes. The more consequential change is quieter. Drones, sensors and battlefield AI have opened a second supply line, one measured in batteries, generators and watts, and the industrial base that would sustain it sits largely in China. The United States Department of Defense is routinely called the world's largest institutional consumer of petroleum. The description traces back to Neta Crawford's 2019 paper for Brown University's Costs of War project, and it is an estimate rather than a measured ranking, since nobody publishes a global league table of institutional fuel buyers. The underlying figures are drawn largely from the years between 2001 and 2017. The statistic nonetheless circulates constantly, usually stripped of its qualifications and usually attached to an argument about carbon. That argument has been had. The development worth attention is that the fuel problem has acquired a companion. During a large US Army exercise in Hawaii in late 2025, electrical energy turned out to be a limiting factor on tactical operations in ways that conventional sustainment planning had not anticipated. Formations retained their vehicles, their weapons and their people, and still lost capability because they could not keep their electronics charged. The commodity nobody counts Armies measure operational reach in commodities. Fuel determines how far a formation can move, ammunition how long it can fight, water and rations how long it can stay in the field. Electricity has never had that status. It hides inside the fuel accounts as diesel for generators, or inside the equipment accounts as the generators themselves. The argument now being made inside the US Army is that this should change, and that electrical energy should be planned, forecast and reported as a commodity in its own right. The proposal, published in Army Sustainment in July, is blunt about why. A commander can usually say how many hours of fuel or rounds of ammunition a unit has left. Very few can say how long that same unit can stay connected, informed and synchronised before its batteries run flat. A battalion command post typically runs a legacy generator continuously to serve a fraction of its rated output, burning fuel inefficiently and radiating heat and noise for no operational return. Infantry carry batteries, and without somewhere to recharge them they end up tethered to the vehicles they were meant to disperse away from. The Ukrainian experience has turned this from a staff college argument into an observed problem. A study for the US Center for Army Lessons Learned, published in March, describes electricity as both a strategic target and a tactical necessity in that war. Radios, drones, command systems, electronic warfare equipment, precision weapons and medical devices all stop working without it. Unlike fuel and ammunition, battlefield electricity has no established supply chain behind it. Batteries deplete quickly, generators announce themselves to anyone with a thermal sight or a scanner, and the fuel convoys that would resupply either are among the most attractive targets on the battlefield. Ukrainian units improvise, mixing military kit with commercial power banks, solar panels and vehicle inverters, because the alternative is silence. The cloud stops where the fibre does Commercial artificial intelligence has solved its own power problem by moving towards the electricity. The IEA now expects data centre consumption to roughly double from around 485 terawatt-hours in 2025 to 950 by 2030, and developers increasingly site and power facilities with grid access as a primary constraint. That option does not exist for a platoon in a treeline. Military AI has to run where the power is scarcest, on the inference budget of a drone, a vehicle or a soldier's back. DARPA has been explicit about the consequences. Its Mapping Machine Learning to Physics programme sets out to build energy-aware machine learning, measuring what performance a model returns for each joule consumed rather than optimising for accuracy alone. The framing example is an uncrewed aircraft, where the total power available has to be shared between propulsion, communications, data processing and mission planning, and where a small loss of model accuracy may be worth accepting in exchange for the range needed to finish the mission. This pulls military and civilian AI in opposite directions. The commercial frontier rewards scale, because scale can be fed. The tactical edge rewards whatever still produces useful answers after the generator has been found, jammed or destroyed. The winning military model may be considerably less capable than the best available model, and win anyway. Watts have a signature Concentrating computing power at the front creates something to aim at. Generators produce heat, noise and electromagnetic emissions. Charging points gather vehicles and people. Processing nodes need cooling. The Army writing on this treats signature reduction as one of the principal arguments for replacing legacy generators with battery and inverter systems, which is a revealing justification: the case for electrification is being made on survivability rather than emissions. Russia has been making the strategic version of the same point for four years. Ukrenergo's chairman said in July that Russian forces now use up to twenty-five drones in a single attack on an energy facility, with strikes on the power system up by more than a third year on year, and that Ukraine has lost roughly half its generating capacity. Ukrainian defence intelligence has since described a purpose-built Russian drone designed to cut power line towers, attacking distribution rather than generation. A force that digitises without hardening its power supply has built a new dependency and helpfully labelled it. Prohibition without a factory The uncomfortable part is industrial. Section 154 of the FY2024 National Defense Authorization Act prohibits the Pentagon, from 1 October 2027, from spending appropriated funds on batteries produced by CATL, BYD, Envision, EVE Energy, Gotion and Hithium. The reasoning is sound. The supply position is not. The IEA records China as accounting for more than 80 per cent of global battery cell production in 2025, with higher shares again in the cathode and anode materials that go into the cells, and nearly all cells worldwide supplied by companies headquartered in China, Korea or Japan. Western militaries reducing their exposure to fuel convoys are therefore walking towards a dependency that is more concentrated than the one they are leaving. Oil has many producers and a deep spot market. Battery cells at scale have very few. A prohibition with no corresponding factory is a plan to be poorer at the thing you have decided matters, on a date you have announced in advance. The strongest objection to all of this is one of proportion. Jet fuel and diesel still account for the overwhelming majority of military energy by any measure, and most battlefield electricity is generated by burning diesel, which means the second supply line is for now a branch of the first rather than a rival to it. The Army officers making the case for a distinct energy commodity are advocates writing in professional journals, arguing for attention and resources, and a service that has managed batteries adequately for decades may simply manage them better without any doctrinal reinvention. It is also possible that efficiency wins quietly, that model compression and better silicon shrink tactical demand faster than demand grows, and that the whole question resolves itself into a procurement footnote. That would be the comfortable outcome. It requires believing that a force which has spent a decade adding sensors, drones, electronic warfare and automated decision support to every echelon will stop adding them. The more plausible reading is that the West has correctly identified electricity as a strategic input, correctly identified its exposure, and responded mainly by restricting purchases rather than by building capacity. Ukraine has already demonstrated what happens when an adversary works out that the power supply is the target. That battlefield electricity matters is settled. The open question for the next European defence review is which democracy intends to manufacture it. Sources: Costs of War Project, Brown University; US Army (Army Sustainment; Center for Army Lessons Learned); DARPA; International Energy Agency; Kyiv Post; The Defense Post; Arnold & Porter analysis of the FY2024 NDAA.