At the World Economic Forum in Davos in January 2026, Elon Musk told BlackRock CEO Larry Fink that full-scale, solar-powered AI data centers in space could be operational within two to three years. One week later, SpaceX announced it was acquiring xAI.
Confirmed
SpaceX filed a request with the US Federal Communications Commission for an "orbital data center" satellite constellation that could include up to one million satellites in low Earth orbit, powered by solar energy and connected via laser links — computing hardware placed in space and run on continuous sunlight, rather than drawing electricity from terrestrial power grids. The underlying physical logic is real and specific: solar panels in low Earth orbit harvest roughly five times more effective solar energy than ground-based panels, according to Musk's own stated figure, because there's no atmosphere to filter the light and satellites orbiting roughly every 90 minutes get far more continuous exposure than any fixed ground installation. Musk framed the motivation directly around a real problem covered in ATD's companion piece on data center energy demand: escalating AI power consumption, one IEA scenario cited by industry press putting data center electricity use as high as 1,000 terawatt-hours by 2026 in a worst-case trajectory, alongside the cooling and water-use conflicts already straining communities hosting terrestrial facilities.
In March 2026, Musk announced the "Terafab factory" project, explicitly tying together SpaceX's orbital satellite constellation, xAI's AI compute demands, and Tesla's manufacturing scale into one initiative. Separately, at least one smaller company has already demonstrated a working proof of concept: Orbit AI's DeStarlink Genesis-1 satellite, launched December 10, 2025, carries Nvidia AI processing hardware powered entirely by space-grade solar panels, performing AI inference operations in orbit without ground-based power. The core enabling technology behind all of this is SpaceX's Starship, whose reusability, if achieved, Musk projects could cut the cost of reaching orbit by roughly a factor of 100, to under $100 per pound — a cost threshold that would make space-based infrastructure of this scale economically plausible for the first time.
Genuinely uncertain
"Two to three years" from a January 2026 statement is Musk's own timeline, and Tesla and SpaceX timelines carry a well-documented history of running later than announced — a caveat that applies here with the same force it applies to Optimus production targets covered in ATD's robotics reporting. Full Starship reusability, the technical foundation the entire cost case depends on, was described by Musk himself as something the company hopes to prove this year, not something already achieved. The FCC has approved an additional 7,500 Starlink satellites but deferred authorization for a further 14,988 proposed satellites, meaning the regulatory pathway for a constellation of the scale being discussed is not yet secured. Whether space-based solar for AI computing becomes genuinely cost-competitive with terrestrial power at scale, versus remaining a niche or symbolic application, is a real open engineering and economic question with credible experts on both sides.
Worth examining honestly
PROMETHEUS would tell you this is Kardashev-scale thinking taken seriously as an engineering roadmap rather than science fiction — a genuine attempt to solve AI's energy and cooling problem by moving the infrastructure somewhere the problem doesn't exist in the same form. GHOST PROTOCOL would add the less romantic read: this is also a vertically integrated bet by one person controlling rockets, AI compute, and manufacturing simultaneously, concentrating an enormous amount of critical infrastructure — and the ability to shape how it's governed — in a single company's hands.
The Door This Opens
This door isn't asking whether space-based AI infrastructure is possible — the physics and the early proof-of-concept say it plausibly is. It's asking whether the world wants its next generation of critical AI computing infrastructure built by whoever can launch the most satellites fastest, and what that concentration of orbital infrastructure means for everyone who doesn't own a rocket company.