Electrons Are the New Ammonia

About half the nitrogen in your body passed through a chemical reactor on its way to you.

The Haber-Bosch process pulls nitrogen out of the air and turns it into ammonia, ammonia becomes fertilizer, and fertilizer feeds most of the people on Earth. The part everyone forgets is what feeds the reactor. Natural gas. And not only as fuel: the gas donates the hydrogen atoms that end up in the ammonia itself. Roughly eighty percent of the production cost of nitrogen fertilizer is the gas. Which means the price of gas quietly sets the price of food, everywhere, and whoever controls the gas holds a lever on every field in the world.

AI is converging on the same shape, and the feedstock is the electron.

Training and inference do not consume electricity the way an office consumes lighting. The kilowatt-hours are the raw material. They set an upper bound on how much intelligence you can produce, and as the models and the tooling standardize, the marginal cost of intelligence collapses toward the marginal cost of energy. Everything else in the stack can be bought, copied, or hired. The gigawatts cannot. They have to exist, physically, next to the machines.

Electricity does not ship well across oceans. Compute ships perfectly. So the workload moves to the power, not the other way around. Cheap, abundant electricity becomes a gravity well: first the data centers come, then the clusters, then the labs that need the clusters, then the people who want to be near the labs, then the models themselves. Whoever owns the cheapest clean gigawatts does not just get lower operating costs. They get, eventually, the center of mass of the whole field.

The standard reply is that this does not matter, because the West controls the advanced chips, and without the chips the gigawatts are just heat. I believed a version of this longer than I should have. The evidence has been arguing against it for years now. China keeps landing close to the frontier without the best silicon. Not at the frontier, close to it, and the distinction matters, but look at what even "close" proves. If export controls on the most advanced chips were a wall, the gap should be growing. It is not. The chips were a brake, not a barrier.

And there is a more uncomfortable reading of the same evidence. What does an engineering culture do when it cannot buy the best hardware? It gets better at everything else. Architecture, training efficiency, squeezing more intelligence out of every watt and every transistor. Scarcity is a harsh but effective teacher, and we may have spent years enrolling our rival in exactly the course that matters most in the era where power is the scarce input. The controls meant to slow them down may have been hardening them for the race we now say matters. I do not claim this as settled fact. I claim it is plausible enough that betting the West's position on the durability of a chip advantage looks reckless.

Because we have seen this film before, and recently, and we were in it.

Before the invasion of Ukraine, Russia supplied close to half of the EU's natural gas imports. It was also the EU's largest single supplier of fertilizer, at somewhere around a quarter to a third of imports. Cheap, convenient, rational every individual year. Then the dependency was weaponized, and here is the detail that should be taught in every strategy course: we could not stop. Fertilizer was deliberately spared from sanctions, out of fear for food security, and EU fertilizer imports from Russia in 2025 were running above their pre-war level. Not reduced. Above. A continent at economic war with its supplier, buying more than before, because the alternative was hunger politics no government would survive. That is what a feedstock dependency looks like from the inside. You see it clearly, and you keep paying.

Now run the same film with electrons and models, and notice the one way it is worse. Fertilizer could at least be stockpiled. You could buy a season ahead while you rebuilt your own capacity. Access to an intelligence trained somewhere else cannot be stockpiled. There is no warehouse for it. If the center of mass moves and your economy runs on models trained in someone else's gravity well, the dependency is real-time, continuous, and cancelable by someone else, and the buffer is zero.

So far this could be mistaken for a text about China. It is not. China's electricity is not cheap because of some unfair trick. It is cheap because they build. Reactors, dams, grids, transmission at a scale and pace the West has chosen not to attempt. Meanwhile expensive electricity in Europe is not a fact of geography. It is the sum of our own decisions: permitting processes measured in decades, nuclear capacity closed or blocked, grids underbuilt, every individual decision defensible and comfortable at the time. Russia did not trick us into the gas dependency either. We walked in because it was cheaper than building ourselves, one comfortable year after another. The pattern is the same and the author of the pattern is us.

There is a serious objection to all of this, and it deserves space. Sovereignty requirements, data residency, latency, and the audit demands of regulated industries all anchor workloads at home. A bank cannot train on infrastructure it cannot answer for, whatever the price per kilowatt-hour. So the realistic failure mode is not that everything moves. It is quieter than that. The price-sensitive bulk of the field migrates to the cheap power, the protected regulated core stays home, and the home market shrinks, year by year, until it is too small to carry its own frontier capability. You do not lose everything. You lose the mass that everything else depended on, and you keep a well-regulated remnant. That is not reassurance. That is the same ending in slow motion.

So the argument, in order, without skipping steps. The evidence: a rival keeps getting close to the frontier without the best chips. The conclusion: the chip moat is shallower than we think, which promotes power to the deciding long-term input. The alarm: the West is bad at building power, on purpose, by accumulated choice, and that, not any rival's strength, is the actual vulnerability.

The AI era has a feedstock, and it is power. Whoever owns the cheap, clean, abundant electricity eventually owns the compute, the labs, and the models, because the workload moves to where the power is, not the other way around. We thought the advanced chips were our moat. A rival that keeps coming close without them has shown the silicon was a brake, not a barrier, and that whoever is forced to solve the hard problem without the best tools gets hardened in exactly the efficiency that counts once power is the scarce commodity. We have seen this film before. We made ourselves dependent on Russian gas and fertilizer, not because we were forced, but because it was cheaper than building, and we could not stop even after the dependency was turned against us. Now the same choice arrives again, with power, and this time there is nothing to stockpile, because you cannot store access to an intelligence trained somewhere else.

The danger is not that China is strong. The danger is that we, one comfortable short-term decision at a time, choose away the ability to stand on our own feet, and call it caution.