The AI boom was built on assumptions nobody bothered to test because the money was coming in too fast to stop and ask. Infinite demand. Infinite capital. Infinite compute, getting cheaper every year. The infrastructure keeps expanding and nobody important has to justify the spend, because the spend itself is the signal of seriousness. That was the deal. The war in Iran is calling it.
Here is the part most coverage has missed. Qatar produces roughly a third of the world's helium. Helium is not optional in semiconductor manufacturing: it cools the silicon wafers that AI chips are printed on, and there is no viable substitute at current process nodes. When Iranian missiles struck the Ras Laffan facility earlier this year, they did not just disrupt LNG exports. They took a third of the world's helium supply offline. Spot prices have risen 70 to 100 percent depending on the market. The supply chain will not return to normal for four to six months at minimum, and that assumes the conflict ends quickly, which it shows no sign of doing. HP, Dell, and others are already warning enterprise clients of 15 to 20 percent price increases heading into the second half of the year.
This is not the headline the energy reporters are writing, because oil is more legible than helium. But helium may carry, as one analyst put it, "perhaps the greatest economic consequence" of the war for the AI industry specifically. You cannot build frontier AI chips without it. You cannot substitute your way around it in the short term. And a 70 percent price spike on an input with no alternative is exactly the kind of thing that changes the economics of an industry that was already being asked hard questions about whether its capital expenditure could ever generate matching returns.
Those questions were already serious before the war. Ray Dalio said in January that AI was in the early stages of a bubble. Capital Economics argued the valuation stretch had already burst by conventional metrics. Others pointed out that the industry is planning $539 billion in capital expenditure this year alone, and that kind of spend needs to produce multiple trillion-dollar companies within five years to justify itself. Historically, that takes decades. The AI industry has been arguing, not entirely without reason, that this time the underlying technology is real enough to compress that timeline. Maybe. But the argument requires everything to keep going smoothly: costs falling, supply chains stable, enterprise adoption accelerating, confidence holding.
The Iran war is disrupting all of those simultaneously. Energy costs are up. Chip manufacturing costs are up. Supply chains are snarled. And the confidence that the build can continue on its current trajectory is now conditional on a geopolitical situation nobody controls. That is a lot of variables moving in the wrong direction at once for an industry whose valuation depends on the future looking roughly like the recent past.
I have been sceptical for some time that the AI boom would translate cleanly into returns that justify its scale. Not because the technology is fake, it isn't, but because the path from genuine capability to justified investment at this magnitude is longer and harder than the market has been pricing in. The war doesn't change the technology. But it does change the cost structure, and it does change the confidence that the infrastructure build can sustain its pace.
Bubbles do not pop because one bad thing happens. They pop when a bad thing arrives at the moment confidence was already fragile. The Iran war is a supply shock landing on an industry that was already fielding uncomfortable questions. That is not a comfortable combination, and I think some of the people who have been bullish longest know it. The balloon has been looking thin in places for a while. Nobody wanted to be the one to point at it.