Data Centers in Space: The Physics That Will Decide If AI Moves to Orbit

Right now, as you read this, an Nvidia H100 graphics card — the very same one that trains the world's large AI models on Earth — is orbiting the planet at more than 27,000 kilometers per hour. The company Starcloud launched it on November 2, 2025, aboard a SpaceX rocket. It fits inside a satellite the size of a small refrigerator and, in December of that year, it became the first computer ever to train a language model outside the atmosphere. This is not science fiction. It is a fact with a date, a registration, and an orbit.

So the question is no longer whether you can put an AI chip in space. You can: it's done and it's running. The question — far harder and far more interesting — is whether it makes sense to send up not a chip, but an entire data center, with its thousands of processors, its electric hunger, and its heat. And that answer is decided, before any visionary and before any investor, by physics. This is an attempt to tell, honestly and without smoke, what is already known, what is in doubt, and what remains unsolved.

The problem starts on the ground

To understand why the industry is looking up, you have to look down first — at the planet's electricity bill. Data centers consumed roughly 536 terawatt-hours in 2025, close to 2% of all the electricity the world produces. The most-cited projections expect that figure to double by 2030, driven almost entirely by AI. In the United States alone, data-center power demand jumps from about 31 gigawatts in 2025 to a projected 66 by 2027 — more than double in two years.

A gigawatt is, roughly, the consumption of a mid-sized city. Multiply that by dozens, and add that the demand concentrates in a few counties whose grids were never designed for such a bite. The consequences are already visible: saturated substations, projects waiting years for a connection, and a growing local backlash against megacenters that compete for water and push up the price of power. AI is running into a limit that isn't about cleverness, but about earthbound infrastructure. And when a resource runs dry in one place, industry does what it has always done: look for it somewhere else. In this case, somewhere else is four hundred kilometers up.

Why space tempts the AI industry

The logic is seductive and, in part, old. The idea of harvesting solar energy in space dates back to the 1970s, when physicist Gerard O'Neill and NASA dreamed of enormous orbital panels beaming electricity down to Earth. That never left the page, because there was no demand to justify the cost. AI has just supplied that demand.

An Earth-bound data center fights for three scarce resources: power, water, and land. In the right orbit — a sun-synchronous one, always facing the Sun — a solar panel receives light almost without interruption and can be, according to figures Google published with its Project Suncatcher, up to eight times more productive than on the ground. No clouds, no night, no seasons, no grid to overload.

Water, too, leaves the equation. On Earth, cooling a processor that dissipates 700 watts demands evaporation towers that drink millions of liters — one of the most repeated environmental charges against the industry. In the vacuum, not a drop is needed: space itself, at less than three degrees above absolute zero, is a theoretically infinite heat sink. That is the promise the two companies leading the race repeat alike.

Who is already building

This is not the lone bet of an eccentric. Starcloud, backed by Nvidia, raised 170 million dollars in March 2026 and is preparing a second satellite that will carry the largest deployable radiator ever flown on a private spacecraft — a hint that its engineers know exactly where the problem lies. Its pioneering satellite didn't just train a small model; it also ran a version of Gemini, Google's model, in orbit, proving consumer hardware survives the trip.

Google, for its part, doesn't want one satellite: it wants a swarm. Its Project Suncatcher describes constellations of 81 satellites loaded with its TPU chips, spaced barely one to two hundred meters apart within a one-kilometer radius, linked by ultra-fast optical connections to work as a single distributed data center in the sky. The first two prototypes, in partnership with Earth-imaging company Planet Labs, are planned for early 2027.

And one figure surprised Google's own engineers. When they bombarded one of their TPU chips with a proton beam mimicking years of cosmic radiation, the processor did not fail until doses far higher than feared. Modern electronics, designed for Earth, withstands the punishment of space considerably better than intuition suggested. An obstacle many considered decisive turned out to be softer than expected.

The wall no one has torn down yet

This is where honest science writing has to part ways with the hype, because the greatest enemy of this idea isn't radiation or cost: it's heat, and it's pure high-school physics. Space is an infinite heat sink, true, but with an essential catch: in the vacuum there is no air. And without air, heat cannot be carried away by convection — the way your computer's fan or a data center's air conditioning expels it. In the vacuum, the only way to shed heat is to radiate it, to emit it as infrared light into the blackness.

And radiating is slow. Very slow. The amount of heat a body can emit grows with its surface area and its temperature, but at the temperatures a chip runs, that emission is modest. Dissipating the hundreds of megawatts — or the gigawatts — a real data center would consume would demand radiators the size of football fields, deployed and precisely oriented in orbit without shadowing one another. A fridge-sized satellite can radiate a chip's 700 watts. No one has yet shown that you can build, fold into a rocket, deploy without failure, and maintain the thermal equivalent of a power plant. That is the leap from the model to the factory, and that leap remains untaken.

And then there's the rocket's bill

The second wall is economic and just as stubborn. Putting a kilogram into orbit costs more than a thousand dollars today, even with the reusable rockets that have cheapened access to space. A large data center would weigh thousands of tons between processors, structure, panels, and radiators: thousands of launches. Google's own analysis is transparent about it and admits the model only closes if launch prices fall below 200 dollars per kilogram — a figure its authors don't expect before the mid-2030s, and one that depends entirely on next-generation rockets like Starship keeping a cost promise they have not yet kept.

More unknowns remain, and they deserve to be stated plainly rather than hidden beneath the shine of a headline. No one knows how to repair a failed server four hundred kilometers up, where no hand and no technician reaches: on Earth, a failed chip is swapped in minutes; in orbit it is, for now, unrecoverable scrap that also feeds the space-junk problem. No one knows for certain how those processors age after years of real radiation and extreme thermal cycling, rather than an afternoon in a particle accelerator. And it is not proven that, adding launch, cooling, maintenance, and replacement, the final math favors space over the more boring and more likely alternative: a sunny desert covered in panels and batteries.

Latency decides who goes up and who stays

There's a technical nuance often lost in the debate that will probably mark the real boundary. Not all AI tasks are equal. Training a model — the slow, massive process of teaching it from trillions of data points — doesn't need an instant answer: it can take weeks, and no one minds if the data travels a few milliseconds farther. It's the ideal load to send where the Sun never sets. Inference, on the other hand — every time you ask an assistant something and wait for the reply — lives or dies by latency: no one wants their question to fly up to orbit and back before it's answered. That boundary, between what tolerates distance and what doesn't, is the one that will decide which part of AI might move and which will stay, forever, close to home.

So, is it the future or is it smoke?

It's both, depending on where you look, and being precise demands holding both at once. As a demonstration, it already happened: there is artificial intelligence training in orbit today, and just two years ago that sounded like a boast. As a replacement for Earth-bound data centers, it remains a hypothesis physics has not yet authorized and economics does not yet support; claiming otherwise would be selling a future no one has built.

The likeliest outcome is neither orbital utopia nor a mocking dismissal, but something in between and gradual: a handful of specific workloads — starting with model training — migrating little by little to where energy is infinite and free, while Earth keeps everything that must stay close to the user. The leap from the lone chip already in orbit to the gigawatt in the sky won't be made by a startup with a good slide deck, nor by a billionaire with a good headline. It will be made, or not made, by the stubborn arithmetic of heat and cost. And that arithmetic, unlike the promises, cannot be accelerated with a funding round. It is being written right now, equation by equation, over our heads.