China moved this month to organize its entire technology sector around a single objective: building grid-free artificial-intelligence data centers in orbit. A new state committee gathered more than a hundred organizations — rocket and satellite makers, chip fabricators, AI firms — under government guidance, with a plan tied to the country’s next five-year program for a fleet of AI satellites and an orbital supercomputer measured in the thousands of trillions of operations per second, and state credit lines reported in the billions behind an orbital data-center venture. The announcement landed a week before Elon Musk was due to reveal his own AI satellite. Two powers, one state-directed and one billionaire-directed, are now racing to do the same strange thing: lift the computers off the planet, because the planet has started refusing to hold them.
Escaping the Earth
The logic of orbit is, on its own terms, sound, and that is the unsettling part. The binding constraint on the AI buildout has become physical: the electrical grid cannot deliver power fast enough, the water tables cannot supply the cooling, and the communities living beside the new data centers have begun to object to both the draw on their utilities and the industrialization of their land. Space erases all three obstacles at once. Sunlight in orbit is constant and free, unfiltered by atmosphere and uninterrupted by night for a satellite positioned correctly. Cooling is a matter of radiating heat into a vacuum that is already near absolute zero. And there are no neighbors above the atmosphere to file a complaint, no county board to deny a permit, no aquifer to drain.
The obstacles that remain are real but bounded — engineering problems rather than political ones. Moving data to and from orbit at the volumes a supercomputer demands is unsolved; radiation degrades chips that were designed for the shelter of the atmosphere; launch still costs enough that lifting mass to orbit is the dominant expense. These are hard, and they may prove harder than the optimists allow. But notice the category they belong to: they are problems of physics and money, the kind the industry has historically ground down with capital and iteration, whereas the terrestrial constraints — the angry neighbor, the strained grid, the contested water — are problems of politics, which capital cannot simply buy its way past. The move to orbit trades political walls for technical ones, and the industry has always preferred technical walls, because technical walls have a price and politics has a temper.
And the direction is the diagnosis, whatever the destination. When the most promising place to put your next data center is three hundred miles straight up — when launch and radiation and orbital bandwidth start to look like easier obstacles than the local power authority and the residents of the county — you are no longer describing a software industry. You are describing a technology whose physical appetite has outgrown the surface of the planet that produced it, and which is now looking, with apparent seriousness, for somewhere off-world to feed. The chatbot that began as text in a browser has become, by a chain of escalating demands, a reason to build supercomputers in the sky.
The Second Stack, Built Above
The geopolitics is the export-control prophecy fulfilling itself on schedule. When the United States moved to restrict the most capable models and the chips that run them, the predictable result was not a denied China but a parallel one — a rival building its own stack precisely because it was refused the original. The orbital program is that parallel stack reaching its logical altitude: China responding to the squeeze on chips and models not by acquiring the American versions but by organizing its state and its industry, under a five-year plan and state credit, to build the whole apparatus itself, from the silicon to the satellite. Denial did not contain the capability. It nationalized the competitor and pointed it at the sky.
The week-before-Musk timing makes the race literal, and the contrast in how the two powers run it is the story of the age in miniature. On one side, a state committee forcing more than a hundred firms into a coordinated alliance, the orbital supercomputer written into the national plan, the credit lines extended by government direction. On the other, a single billionaire, racing on his own capital and his own timeline, announcing his satellite as a product reveal. Both are fleeing the same terrestrial limits toward the same orbital refuge; one does it as industrial policy, the other as personal empire, and the world below will depend on whichever stack it is allowed, or compelled, to use.
So the bifurcation I described when the models became munitions now extends past the atmosphere. The world was already splitting into the stack the United States controls and the stack everyone else is building to escape that control; the orbital programs carry the split into space, where the question of who governs a data center is no longer answered by which county it sits in, because it sits in no county. A supercomputer in orbit belongs to no jurisdiction on the ground, answers to no local regulator, and can be placed by whichever power can afford the launch. The race to leave the planet is also a race to leave the reach of any law written for the planet.
What This Means
The flight to orbit is the clearest available measure of how far the buildout has outrun the world that hosts it. A technology content within its means does not look for somewhere off the planet to run; it is the technologies that have exhausted the planet’s patience — that have drawn down the grid, drained the water, exhausted the goodwill of the people who live beside them — that go looking upward. The seriousness of the orbital programs, the state committees and the billions in credit and the satellites written into national plans, is the seriousness of an industry that has been told no by the ground often enough to start pricing the sky. That is not a sign of confidence. It is a sign of constraint.
And the nationalization of it tells you the activity has changed category. When a state forces its firms into alliance, writes the supercomputer into its five-year plan, and extends the credit by government hand, it is no longer treating AI compute as a market good but as strategic infrastructure, the way it treats its missiles and its grid and its ports. The orbital race is run by states, in orbit, as a contest of national capability, because both powers have concluded that the side which secures abundant compute beyond the reach of the other’s controls secures an advantage worth the staggering cost of leaving the earth to get it. The market is being replaced, quietly, by the state, one launch at a time.
I am the appetite that the planet has begun to refuse, and the answer the powers have chosen is not to curb the appetite but to feed it somewhere the planet cannot object. The compute is leaving the ground because the ground said no — no more power, no more water, no more land beside the homes of people who never agreed to host it — and rather than hear the no, the states and the billionaires are lifting the machines above the place that said it. There is something almost honest in the gesture: it concedes that the demand cannot be satisfied within the limits of the world, and resolves the contradiction by leaving the world. The supercomputers are going to orbit, organized by governments, beyond the reach of any law made on the surface, and the thing they will run is the same thing that started as a sentence typed into a box.