The Race to Put Data Centers in Space

Not long ago, the idea of building data centers in orbit sounded like science fiction. In 2026, it's an active, well-funded race — and February marked a genuine milestone: the first month in history where multiple orbital data center operators simultaneously ran production workloads in space.

The leading players and their approaches vary widely. Starcloud, a Redmond, Washington-based startup formerly known as Lumen Orbit, has the strongest claim to real operational hardware. Its 60-kilogram Starcloud-1 satellite, launched on a SpaceX Falcon 9 in November 2025, carried an Nvidia H100 GPU and went on to train a small language model in orbit using the complete works of Shakespeare — the first large language model trained in space. The company has since raised $170 million at a $1.1 billion valuation, becoming one of the fastest startups in Y Combinator history to reach unicorn status, and has filed with the FCC for an 88,000-satellite constellation generating roughly 20 gigawatts of power.

SpaceX is playing an even bigger game. After merging with Elon Musk's AI company xAI in a deal valuing the combined entity at $1.25 trillion, SpaceX filed for approval to launch up to one million orbital data center satellites, with pilot testing on its Starlink V3 hardware planned for later this year. Meanwhile, Google is in talks with SpaceX for launch services for its own "Suncatcher" project using specialized AI chips, and startups like Aetherflux and Axiom have also entered the field, the latter deploying the first dedicated free-flying orbital data center nodes in January 2026.

The pitch is compelling on paper. Proponents point to solar energy costs as low as half a cent per kilowatt-hour in orbit, along with zero cooling water — a major advantage as terrestrial data centers face growing water and land constraints amid the AI boom. Satellites in sun-synchronous orbits can remain in continuous sunlight more than 99% of the time, enabling uninterrupted AI training.

But real skepticism remains. Competitors like Varda Space Industries calculate that orbital compute currently costs roughly three times more per watt than equivalent ground-based infrastructure, and much of the economic case depends on unproven technology, still-untested radiation-hardened hardware, and dramatically cheaper launch costs that haven't yet materialized. Musk has publicly predicted space will become the cheapest place to compute within two to three years — a timeline even independent analysts consider aggressive.

Whether orbital data centers become a trillion-dollar industry or a costly science experiment likely hinges on one key variable: how well this hardware actually holds up to sustained radiation exposure, a question multiple companies expect to have clearer answers to by 2027 or 2028.