Pieza independiente
The metabolism
Movement 2
1942–2026
I
On 2 December 1942, on a squash court under the west stands of a football stadium closed by the war, in the middle of the University of Chicago campus, forty-nine people stood watching a pile of black bricks. Forty-nine is the count that gets repeated. Some lists say forty-eight [nobody has managed to settle it].
Four hundred tonnes of graphite, six tonnes of uranium metal, fifty of oxide, fifty-seven layers stacked by hand by students who finished their shift black to the eyebrows. Nobody had ever built the thing. There was no shielding, no containment, no cooling, and no evacuation plan for the three and a half million people living around it.
George Weil was down below, alone, with his hand on the last cadmium rod. He drew it out fifteen centimetres at a time, and each time Enrico Fermi did a sum on his slide rule and compared the answer to the clicking of the counters that Leona Woods was reading, twenty-three years old, the only woman in the room.
At 15:25 the counters stopped climbing in steps and started climbing on their own.
Fermi let it run for twenty-eight minutes and ordered it shut down. Arthur Compton telephoned Harvard with a code he had invented on the spot. “The Italian navigator has just landed in the new world,” he said. The voice at the other end asked how the natives were, and Compton answered that everyone had arrived safe and happy.
The first self-sustaining chain reaction in history produced half a watt.
Half a watt lights no bulb. It warms no finger. All that afternoon produced was a threshold crossed and a date: the moment a number everyone called k reached one and stayed there.
II
The safety system was a man with an axe.
His name was Norman Hilberry, he stood on top of the pile, and his job was to cut a rope holding a cadmium rod that would drop into the pile by gravity if anything went wrong. There were also three volunteers above with buckets of cadmium salt solution, to throw over it. They were called the suicide squad.
One word survives from that episode and is still used in every power station in the world to mean emergency shutdown. Its origin is disputed, and anyone who explains it with too much confidence is embroidering. Hilberry did say, years later, that the axe was given to him.
Hold on to that image, because the end of all this comes back to it. The ultimate guarantee of the first nuclear machine was a man with a name, a salary and a shift, physically present, holding a hand tool.
III
I went to look at a substation at twenty to five on a Tuesday morning in July, because at that hour there is nobody there and because the place does not take visitors.
Three degrees. July in the wrong hemisphere. The smell is not ozone, which is what you expect from having read too much; the smell is hot mineral oil and grass burnt by the herbicide they use to keep the compound bare. The transformers hum at the second harmonic of the grid frequency, a hundred hertz where I was standing [fifty on the grid, double that in the hum], a flat sound that comes from no particular point and after ten minutes settles in behind your eyes.
I counted twenty-two pylons on the transmission line before it disappeared over the rise. On the far side of the fence there were four long windowless halls with the cooling units on the roof forming a ridge. The company that buys that electricity appears on no sign. The name on the gate belongs to a holding company registered in another state.
Two weeks later I got inside a similar hall, with an escort and an eleven-page confidentiality agreement [which I signed without reading it all, like everyone else].
Inside you wear plugs. In the cold aisle it is twenty-two degrees and you can talk; in the hot aisle it goes past thirty-five and the air tastes of dust toasted on metal. The fan noise is broadband, it does not bother you at first, and it leaves a ringing that in my case lasted until mid-afternoon. Nobody works in there. You go in to change something and you come out.
What impressed me most was not the hall. It was the yard at the back, where the generator sets stood in a row with their tanks, and the technician told me how many hours they hold at full load and what the monthly start-up test costs. That figure, which I cannot reproduce here because it was in the agreement, was more than my rent.
IV
What changed in these twenty-seven months was not the power of any model. It was the unit of measure.
Until 2024 the board was counted in parameters. In July 2025 a single infrastructure programme added four and a half gigawatts to what it already had, passed five in development and projected more than two million chips. In September that year a silicon maker and a lab announced a deployment of at least ten gigawatts of systems. In October, another ten of in-house accelerators with a third partner.
A parameter gets announced. A gigawatt has to be manufactured, and manufacturing one means a specific and extremely dull list: power transformers with lead times measured in years, rights of way over land belonging to people who vote, a substation, an evacuation line, cooling water with its discharge permit, and concrete. A great deal of concrete.
The bottleneck is not where the announcement puts it either. It is in the grid interconnection queue, in the power transformer ordered three years ahead, and in the permit to discharge hot water. A campus is delayed by a town council, by a plot whose owner will not sell, and by a piece of steel and copper made by four companies in the world. None of that appears in an earnings call, because none of it goes faster if you pay more.
That is the change of regime. The cognitive layer can be copied in an afternoon; a model's weights fit on a disk that goes in a backpack. A substation cannot be pirated, cannot be exported and cannot be downloaded.
Energy is the only layer of the board whose claims a stranger can audit with a car. You drive there, you look at the fence, and you count whether there are transformers or a vacant lot with a promise nailed to a post.
V
Of all the milestones I logged in two years, one has a property none of the others has. It is binary.
A reactor is critical or it is not. There is no eighty per cent of criticality, no criticality in beta, no phased criticality subject to availability. The physical threshold is the same one crossed that afternoon on the squash court, and crossing it leaves a date somebody has to sign.
In June 2026 two reactors in a federal programme went critical. On 3 September 2026 a microreactor did [another company, another design, a test facility] and the announcement came out on the 16th. Thirteen days between the physical fact and its communication, which is the most honest interval I have found anywhere on the board.
For a month I worked with the idea that this was the test: criticality or silence. It was wrong, and it was wrong for the same reason that makes the afternoon of 1942 beautiful.
Criticality is not electricity. The Chicago pile went critical and gave half a watt. The first time a reactor lit anything it was four light bulbs in an Idaho desert, in December 1951, nine years later. Between crossing the threshold and selling a kilowatt-hour there is a distance measured in decades and in courtrooms.
So the test has to be hardened one more notch, and the one that holds is the supply contract. In September 2026 a geothermal one was signed for three hundred and ninety-six megawatts to feed data centres. It is not an announcement of capacity. It is an obligation to deliver electrons at a price, with a penalty if they do not arrive.
VI
Here I have to say what does not fit, because leaving it out turns everything above into a comfortable thesis.
Two different ledgers of these same twenty-seven months disagree on fusion. One logs four events — construction started, a conditional permit, a measurable deuterium-tritium demonstration per the company itself, a funding round; the other, stricter about what it admits, logs none in twenty-seven months. Both are looking at the same world.
The discrepancy is not an error by one of them. It is the exact map of the border between what leaves concrete and what leaves a press release. Where the density of announcement is highest, the physical work is thinnest, and the reverse: the microreactors that crossed the threshold this year did it almost without headlines, in test facilities, with signed records.
Against myself: the gigawatt also gets announced. Between stated capacity and an energised substation there is a project mortality rate nobody publishes in their earnings deck, and I have written four paragraphs using capacity figures as if they were steel. The way not to cheat is easy to say and irritating to apply. You log the date of the announcement and leave the box open. You close it the day there is a signed record, and not before.
VII
Back to the squash court, where there is something you do not see the first time.
That pile was not built to produce energy. It was built to prove the process worked, and the product that really mattered was not the heat but the plutonium. The first criticality in history was an instrument of knowledge and, above all, an instrument of permission. It existed so that a committee could authorise the next thing.
The reactors that crossed the threshold in 2026 do the same job. They are permission events. The plant that will matter in ten years is the one that turns up on somebody's bill.
Half a watt, twenty-eight minutes, forty-nine people watching. What separates that afternoon from these two years is not the ambition, which was identical, nor the scale, which was ridiculous then.
It is the safety system. In 1942 it was a man standing on top of the machine with an axe, whose name was Norman Hilberry and who could cut the rope in a second and a half.
Today the emergency stop is a contract clause.