When SpaceX filed to go public in 2026, the prospectus described a total addressable market of $28.5 trillion. About 90% of that was attributed not to launch or to Starlink, but to artificial intelligence through xAI, the AI company SpaceX acquired in February 2026. And one of the more striking parts of that AI story is the plan to run AI computing in orbit.
Days after the xAI deal, SpaceX filed plans with the US Federal Communications Commission for a constellation of up to one million "orbital data center" satellites, operating at altitudes between 500 and 2,000 km. That's a staggering number, on the order of a hundred times the size of today's Starlink constellation. Whether or not anything close to it gets built, it's worth understanding why a rocket company thinks the future of computing is above the atmosphere.
Why Anyone Would Put a Data Center in Space
AI is extraordinarily hungry for electricity. Training and running large models requires vast numbers of chips, and those chips need power and cooling around the clock. On Earth, that's increasingly a problem. New data centers compete for grid capacity, water for cooling, and land, and many communities have pushed back against having them built nearby.
Orbit offers one resource in apparently limitless supply: sunlight. Above the atmosphere there are no clouds and no weather, and in the right orbit a satellite can stay in sunshine almost continuously. There's no land to buy, no local water supply to drain, no neighbours to object to noise, and no zoning board. For a company that also owns the world's cheapest heavy launch vehicle, the logic is obvious: if the cost of getting hardware to orbit keeps falling, energy in space could eventually become cheaper than energy on the ground.
SpaceX has already announced a first design, which it calls the AI1 compute satellite. It's an early step, not a finished product.
The Physics Working Against It
The case for orbital data centers is simple. The case against it is subtle, and most of it comes down to heat.
Cooling is harder in space, not easier. It seems intuitive that space, at a few degrees above absolute zero, would be a perfect place to cool computers. It isn't. On Earth, data centers dump heat by blowing air or pumping liquid across hot components, and that heat is carried away by the surrounding air or water. In a vacuum there is nothing to carry heat away. The only route out is radiation: glowing infrared off a surface, which is a slow process. Engineers studying the idea estimate that shedding 10 megawatts of waste heat would take radiator panels roughly the area of two football fields, on top of the solar arrays needed to generate the power in the first place.
Radiation damages chips. Outside the atmosphere, electronics are constantly struck by energetic particles that can flip bits, corrupt calculations, and degrade hardware over time. Space-grade chips are hardened against this, but they typically lag years behind the cutting-edge processors AI companies want. Using ordinary commercial chips in orbit means accepting more errors and faster wear.
You can't swap a failed server. On Earth, data center operators replace hardware every three to five years as better chips arrive, and technicians fix failures daily. In orbit, repairs are extremely expensive or impossible. A satellite full of chips that become obsolete in a few years is a satellite that loses value quickly, and in the AI industry, where chip performance improves fast, obsolescence is the rule.
It all has to be assembled in orbit. Anything the size of a real data center can't be launched in one piece. The panels, radiators, and computing modules would need to be assembled in space, which requires servicing and construction capabilities that are still largely experimental.
Engineering researchers writing about SpaceX's AI1 design in 2026 estimated that a single such satellite is roughly 100 to 1,000 times less capable than a current large data center on Earth. Matching one ground facility would take hundreds of satellites. Matching the AI industry's growth would take far more.
What It Would Actually Be Good For
Not every computing job suits orbit. Anything that needs an instant response is a poor fit, including interactive AI chat, financial trading, and most everyday cloud services, because data has to travel up to the satellite and back down, often through several relay links.
The more plausible early uses are jobs where the data already lives in space, or where speed matters less than raw computation. Processing images from Earth-observation satellites in orbit, instead of beaming raw data to the ground first, is a natural fit. Military and intelligence processing is another. So are long-running AI training jobs that don't need to answer anyone in real time. In other words, the first orbital data centers are likely to serve customers in space before they compete with Amazon, Microsoft, and Google on the ground.
The Space Junk Question
A million satellites raises an obvious concern for anyone who read our piece on space debris. Low Earth orbit is a finite, shared environment, and every large structure in it is both a potential victim of debris strikes and a potential source of new debris. Large data-center satellites with sprawling solar arrays and radiators present big targets. Even with active collision avoidance and reliable deorbiting, a constellation on this scale would change the character of orbit permanently, which is why regulators and astronomers are watching the FCC filing closely.
Thinking about SPCX? Orbital computing is a large part of the growth story behind SpaceX's valuation. Our plain-English SPCX guide covers what you actually own, including the xAI business, and the risks. Not investment advice.
Why This Matters for Investors
At its IPO, SpaceX was valued at around $1.77 trillion. Starlink is profitable today, but much of the gap between what SpaceX earns and what it was valued at rests on AI, including the idea of computing in orbit. Independent analysts were sharply divided: some placed fair value well below the IPO price, arguing the AI assumptions were too optimistic, while the underwriting banks projected revenue growing many times over by 2030.
Orbital data centers are a useful test of that debate. If SpaceX can drive launch costs low enough, solve heat rejection at scale, and find customers willing to pay for computing in orbit, the bull case gains real support. If the physics and economics prove as stubborn as many engineers expect, a meaningful part of the valuation story will need rethinking.
The Bottom Line
Putting AI in orbit isn't science fiction. The pieces exist: cheap launch, abundant solar power, laser links between satellites, and a company with the capital to try. But there is a vast difference between launching satellites and operating industrial-scale computing in space. Heat, radiation, repair, and debris are all hard problems, and none of them get easier just because the idea is exciting.
The realistic outlook is a slow start: small, specialised orbital computing serving space-based customers first, with any challenge to ground-based data centers much further off. Watch whether SpaceX actually launches operational compute satellites, and what they are used for. That will tell you far more than any number in a prospectus.