The Iran War Showed Us AI Has a Body, and It’s Fragile.

We have a habit of talking about artificial intelligence as though it were weightless. It lives in “the cloud.” It runs on “compute.” The language is deliberately ethereal, and it lets us imagine AI as something floating serenely above the messy physical world, a mind without a body. The Iran war of 2026 was, among the many terrible things it was, a brutal correction to that fantasy. It showed, with missiles and mines and jammed satellites, that AI has a body after all, and that the body is made of surprisingly few, surprisingly fragile, surprisingly concentrated things. For anyone whose organisation now depends on AI, and that is nearly everyone, the war is worth reading not as a Middle East story but as an X-ray of what our technology actually rests on.

The war ran at machine speed, and that was the least surprising part

Start with the obvious layer, because it’s the one that made the headlines. This was the most AI-saturated conflict in history, and it barely bothered to hide it. In the opening twelve hours, US and Israel forces struck close to 900 targets; over the first four days, reportedly more than 4,000. That tempo is not humanly possible the old way. It runs on AI-assisted targeting pipelines, systems like Palantir’s and the machine-learning architectures descended from Project Maven, fusing satellite imagery, signals intercepts, and sensor feeds into targets faster than any room full of analysts could. Both sides played. Iran ran drone and missile swarms and poured effort into cognitive warfare, deepfakes and information operations designed to fracture its adversaries’ morale. Commercial space companies provided unjammable low-orbit communications that kept autonomous drone swarms flying through the heaviest electronic warfare ever recorded.

This is the part everyone expected, and it is genuinely significant, the speed and the machine-mediated targeting raise hard questions, sharpened when a strike on a girls’ school in Minab killed scores of children and US senators wrote to ask whether AI had helped choose the target. That question deserves its own reckoning. But the deeper story of this war, the one that should change how a technology leader thinks, is not what AI did to the battlefield. It’s what the battlefield did to AI.

Pull the thread and you find AI’s body, made of things you never think about

Here is where it gets uncomfortable, and genuinely revealing. Iran could not strike an American data centre. But it turned out it didn’t need to, because the physical supply chain underneath AI runs straight through the water it controls.

Take helium. Not a substance most people associate with artificial intelligence, and that’s exactly the point. Liquid helium cools the machines that manufacture semiconductors, the chips every AI system runs on. Qatar produces around 38 percent of the world’s supply, and it ships it out through the Strait of Hormuz. When Iran’s strikes and the Hormuz disruption choked Qatari exports, they put a hand around the throat of semiconductor manufacturing itself, thousands of miles from any battlefield. A war in the Gulf reached out and touched the chip supply of the entire AI industry, through a coolant almost nobody in that industry had ever thought about.

Then energy. AI is astonishingly hungry for electricity, data-centre demand is climbing toward levels that strain national grids, and in the United States a large share of that power comes from gas. Iranian strikes on Qatari gas facilities and the Hormuz squeeze drove gas prices up by around half. The line from a missile in the Persian Gulf to the electricity bill of an American data centre is shorter and straighter than anyone building those data centres wanted to admit.

Then the cables. We picture the internet as borderless, but it is anchored to the ocean floor, and at least seven major submarine communications cables thread across the seabed of the Strait of Hormuz, carrying the cloud services, AI infrastructure, and financial traffic of the entire Gulf. Iran, with mines, midget submarines, and combat divers, sits astride them. The most ethereal-sounding thing we own turns out to rest, literally, on a contested seabed in one of the most volatile stretches of water on earth.

Add the GPS spoofing, over a thousand ships reporting false positions within a day, the largest such campaign ever recorded, and the Iran-linked intrusions into water utilities, healthcare, and industrial control systems, and a picture assembles. The war didn’t have to attack AI directly. It attacked the strait, the helium, the gas, the cable, the signal, and AI shuddered, because those unglamorous physical things are what AI is actually made of.

The chokepoints are few, and that’s the whole problem

If you’ve read my writing before, you’ll know where this lands, because it’s the same pattern that shows up everywhere once you learn to see it: concentration. The reason a regional war could reach the global AI industry is that the physical substrate of AI is astonishingly concentrated. One strait carries a huge share of the world’s oil, gas, helium, and a rack of critical cables. A handful of countries make the advanced chips. A small number of firms run the clouds. Helium comes from very few places. Each of these is a single point of failure, and the war lit them up one after another like a diagnostic test of the whole system.

This is the same lesson the central banks have been warning about in the financial wiring of AI, and the same lesson cyber security teaches about shared dependencies, arriving now through the barrel of an actual war. A system is only as resilient as its narrowest chokepoint, and the AI age has quietly routed itself through a small number of very narrow ones. We built something we describe as limitless on a foundation that is anything but, and we mostly did it without noticing the foundation was there.

The twist: the rivals are standing on the same floor

Here is the detail that should stay with you longest. The United States and China are locked in a bitter contest over who will lead in AI, and the whole world frames that rivalry as a race. But the war exposed something the race narrative hides: at the bottom of the stack, the two rivals depend on many of the same fragile things. The same chokepoints. The same handful of chip fabs. The same scarce inputs. The same physics of coolant and power and undersea glass. China watched this war closely, one Chinese defence assessment reportedly treated it as a live case study in cyber-enabled and cognitive warfare, precisely because Beijing could see its own vulnerabilities reflected in Tehran’s predicament. Rivalry is a story about the top of the stack, who has the best models, the most compute, the leading labs. The war was a story about the bottom of the stack, and at the bottom, the competitors share a floor, and the floor is thin.

What a leader should take from a war they didn’t fight

You are not going to secure the Strait of Hormuz or the global helium trade. So why does this matter to someone running technology or security in an ordinary organisation? Because the war is a vivid, large-scale demonstration of a question you can and should ask about your own AI dependencies, one level down.

The lesson is not “AI is dangerous.” It’s that AI, which you experience as a weightless service you summon on demand, rests on a physical and geographic supply chain that is concentrated, exposed, and almost entirely outside your visibility or control. Most organisations adopting AI have thought carefully about the model and not at all about what the model stands on, the compute, the chips, the power, the handful of providers, the physical geography of it all. The war is the extreme version of a risk that exists, in miniature, in every AI deployment: your capability depends on a narrow base you did not build, cannot see all of, and could lose to an event on the other side of the world.

So ask the questions the war asks, scaled to your own house. What does your AI actually depend on beneath the interface? How many providers, how many suppliers, how much of it traces back to a single region or a single point that, if disrupted, takes you down with it? Where is your fallback if the base fails? These are not exotic geopolitical concerns. They are the same resilience questions we should always have asked about anything critical, made newly urgent by a war that showed the whole world, in the space of a few weeks, exactly how physical the cloud really is.

We called it the cloud because it felt like one. The Iran war was the reminder that clouds, real ones, still need the earth to rain on, and that the earth, right now, is a very contested place.