Related: SpaceX Pulls Off the Largest IPO in History
Elon Musk has confirmed the name of SpaceX’s most ambitious project yet, an artificial intelligence satellite constellation that will be called Starmind. The confirmation puts a label on a plan SpaceX set in motion earlier in 2026, when it filed a request with the Federal Communications Commission to launch and operate a constellation of up to one million satellites designed to run AI computing in space rather than on the ground.
The idea behind Starmind is to move the heavy work of artificial intelligence off Earth and into orbit. Each satellite is planned to carry onboard processors that run AI inference, the work of generating answers from a trained model, and then beam the results back to Earth, bypassing the terrestrial data centers that currently handle that load. Musk has described the satellites as carrying racks of compute connected to one another through inter-satellite laser links, in effect stitching a fleet of spacecraft into a single distributed data center circling the planet.
Power is the part that makes the concept audacious. According to the plan, each satellite would generate as much as 150 kilowatts at peak and around 120 kilowatts consistently from large solar arrays, energy that is far easier to harvest in orbit, where sunlight is constant and uninterrupted by weather or night. Running compute in space sidesteps two of the biggest constraints on AI growth on the ground, the strain that data centers place on electrical grids and the water and land they consume, and it is that pitch, abundant solar power and no terrestrial footprint, that SpaceX is using to justify the scale.
The orbital design is built for flexibility. The satellites are planned to operate between 500 and 2,000 kilometers in altitude across a range of inclinations, from 30 degrees to sun-synchronous orbits, with different clusters spaced at roughly 50-kilometer intervals. That layering would let SpaceX match different groups of satellites to different jobs, placing latency-sensitive work in one shell and heavier batch processing in another, much as a ground operator assigns workloads to different facilities.
The numbers involved are staggering even by SpaceX standards. A million-satellite constellation would dwarf Starlink, the company’s satellite internet network that already counts thousands of spacecraft, and it would require a launch cadence and manufacturing pace that no company has demonstrated. SpaceX has asked the FCC for a waiver of the usual milestone rules, which normally require an operator to deploy half of a constellation within six years of authorization and the full system within nine, a sign of how far beyond the standard regulatory framework the plan reaches. The agency opened the filing for public comment in February.
The timeline starts small. Two prototype satellites, referred to as AI1, are scheduled to launch in early 2027, with volume production targeted by the end of that year. Those first spacecraft will test whether the core ideas hold up, whether the onboard compute survives the radiation and thermal swings of orbit, whether the laser links can move data fast enough between satellites, and whether the economics of launching compute into space can ever compete with building a data center on the ground.
Starmind also fits a larger pattern in Musk’s empire, where SpaceX and his artificial intelligence work have moved closer together. The company has folded AI ambitions into its core business, and an orbital compute layer would give those ambitions a piece of infrastructure no rival could easily match. Whether Starmind becomes a working network or remains a bold filing depends on enormous open questions of cost, heat, and launch capacity, but giving it a name signals that SpaceX intends to treat the AI satellite constellation as a real program rather than a thought experiment. The coming year, and those first AI1 satellites, will show how much of the vision can survive contact with orbit.
Related: SpaceX’s rare Starshield launch · the Tesla–SpaceX merger talk