In February 2026, Musk merged SpaceX with xAI into one company, explicitly to combine xAI's models and compute needs (including the Colossus 1 supercomputer in Memphis, 220,000 GPUs) with SpaceX's launch and satellite expertise. SpaceX filed an application with the FCC in late January for a constellation of up to one million satellites that would serve as an orbital AI data center.
In March, Musk showed an illustration of an "AI Sat Mini" satellite that would provide 100 kilowatts of power for onboard AI processors, dominated by large solar arrays and shown towering over a Starship V3 for scale.
Then on June 8, timed to SpaceX's expected IPO, SpaceX unveiled AI1, the first physical satellite in the constellation, with a 70-meter wingspan — wider than a Boeing 747-8. Musk described it as essentially a large solar panel attached to a computing rack and a radiator, deliberately simpler than a Starlink satellite.
The core argument Musk is making: "As soon as the cost to orbit drops to a low number, it immediately makes extremely compelling sense to put AI in space," and he's claimed space-based, solar-powered data centers will be more cost-effective than terrestrial ones within two to three years, arguing that power on Earth gets harder and more expensive over time while in space it gets cheaper and easier.
Why space, specifically:
Continuous sunlight in the right orbit (no clouds, no night in sun-synchronous orbits)
No land-use or grid-capacity fights with local communities, which are already a major source of pushback for terrestrial data centers over power demand, water use for cooling, and land use
Vacuum offers passive radiative cooling instead of water-hungry cooling systems
The real obstacles nobody's fully solved yet:
Radiative cooling in vacuum is much less efficient per unit area than air/water cooling on Earth — this is probably the biggest physics problem, not the solar power itself
Radiation hardening for GPUs, which weren't designed for space
Laser inter-satellite links for enough bandwidth to make a distributed cluster actually function as one system
Launch cost, even with Starship, at the scale of "a million satellites"
No in-orbit servicing/repair economics yet
So: not a stretch of imagination — it's an active program with hardware already in orbit. Whether the economics pencil out the way Musk claims (2-3 years to cost parity) is the genuinely contested part; that's a forecast, not a settled fact, and Musk has a track record of optimistic timelines on hard engineering problems (Full Self-Driving, Starship's early schedule, Mars colonization dates).