SpaceX has announced the design of its AI1 Compute Satellite, an orbital data center spacecraft that is 100 to 1,000 times less capable than the Earth-based data centers it would eventually need to replace. That gap, laid out by engineering professors Sven Bilén and Wangda Zuo in an analysis for The Conversation, frames the central question of the orbital computing race: not whether companies can launch servers, but whether the economics of heat, hardware refresh and repair can ever close.

The commercial logic is straightforward. AI demand for computing power is growing fast, and Earth-based facilities face land costs, water consumption, local power grid limits and community opposition. In orbit, the sun always shines and no clouds block the panels. There is no land to buy and no zoning board to confront. For a company positioning itself as the railroad, utility and cloud provider of the space economy — the bet behind SpaceX's long-anticipated IPO — orbital data centers are the natural next product line.

What breaks first?

Cooling. Almost all electricity consumed by servers becomes heat, and on Earth that heat leaves through air handlers, chillers, cooling towers and, increasingly, liquid cooling. In space there is no air. Heat can escape only as infrared radiation, a slow process. The professors calculate that removing 10 megawatts of waste heat requires radiator surfaces comparable to two football fields — surface area that comes on top of the solar panels themselves.

Power supply carries its own limits. Even the best solar cells available today convert only about half the sunlight that hits them, and depending on the orbit, Earth shadows the panels for part of each pass. The result is a facility that must be assembled in orbit from thousands of launched components, using in-space assembly equipment that does not yet exist at industrial scale.

Can the hardware keep up?

Server refresh cycles pose the harder economic problem. Operators on Earth replace or upgrade hardware every three to five years as chips improve. In orbit, repair and upgrade become difficult or prohibitively expensive. A computing platform that cannot be updated may go obsolete long before its surrounding infrastructure — solar arrays, radiators, structure — reaches end of life.

Radiation and thermal cycling compound the risk. The hardware operates in near vacuum under constant radiation, swinging from hot in sunlight to cold in shadow many times a day. Orbital debris and micrometeorites can puncture the facility, and a worst-case collision destroys the asset while creating more debris. Even the launch cadence itself draws resistance: SpaceX has faced protests at its Boca Chica, Texas, complex from activists who argue its testing damages the surrounding environment.

Who buys orbital compute first?

Not the mainstream cloud market. Financial transactions, interactive AI services and most cloud applications are extremely sensitive to delay, and data would have to move by radio or laser between Earth and orbit — feasible, as Starlink and Amazon Leo constellations have shown, but at volumes that would balloon. The professors argue the first viable customers are space-based ones: processing Earth observation data, military and intelligence workloads, scientific computing for missions, and specialized compute for satellites.

In plain terms, the satellite serves the constellation before it serves the cloud. That is a real market, but it is measured against a design that currently sits two to three orders of magnitude below terrestrial capacity.

Companies are moving forward regardless of the gap. Whether SpaceX and its rivals can shrink the cooling, refresh and assembly penalties fast enough to reach competitive parity with Earth-based facilities — or whether orbital compute remains a niche service for space assets — will decide if this market exists at all.