Elon Musk folded his artificial intelligence startup xAI into SpaceX early this year in a deal that valued the combined entity at roughly $1.25 trillion, creating what the companies frame as a vertically integrated machine for building data centers in orbit. The logic is straightforward: terrestrial AI clusters are choking on power constraints and cooling costs, and Musk is betting that a rocket company with a satellite network can solve both by moving the computers to where the sun never sets and the vacuum does the thermal work for free.
The merged operation has already branded the effort Starmind. Its first hardware, a satellite designated AI1, spans about 70 meters tip to tip, wider than a Boeing 747, and delivers around 150 kilowatts of peak computing power running xAI's Grok models. SpaceX's roadmap calls for two prototype AI1 satellites to launch in early 2027, with production ramping toward roughly 1 gigawatt of orbital compute per year by late 2027 and commercial service potentially starting in 2028. The company has asked regulators for permission to eventually field up to 1 million such satellites, launched 30 to 50 at a time on Starship.
The pitch for space-based compute rests on two advantages that sound almost too convenient. Solar panels in orbit catch sunlight nearly around the clock without cloud cover or grid interconnection queues, and the cold vacuum lets satellites radiate heat directly without the enormous water and electricity demands of ground-based cooling. No land to acquire, no local utility to negotiate with, no neighbors to placate. Fusing xAI's models with SpaceX's launch, satellite, and Starlink networking expertise under one roof is the structural argument for why this might actually ship before anyone else tries.
The honest counterweight is that most of this remains a dream. Dissipating meaningful heat in a vacuum is genuinely difficult, radiation degrades electronics faster than any terrestrial environment, and there is no technician to dispatch when a server fails 300 miles up. The costs are staggering, and it is worth noting that xAI was burning billions of dollars and needed SpaceX's deep pockets, a financial necessity that drove the merger at least as much as any orbital vision. Timelines this ambitious have a habit of slipping.
For anyone watching SpaceX, orbital data centers are best understood as a long-shot call option stacked atop the real business: launch and Starlink. If Starmind works, it could open a vast new market and justify the towering valuation. If it doesn't, SpaceX still has its core franchise. The space-compute dream is an upside to hope for, not a thesis to buy on its own.
What to watch next is whether the prototype AI1 satellites actually reach orbit on schedule in early 2027 and whether they can demonstrate sustained compute performance without thermal throttling or radiation-induced errors. The regulatory filing for a million-satellite constellation will also test how seriously spectrum authorities take the concept. Until hardware flies and survives, Starmind remains a very expensive PowerPoint slide attached to a very real rocket company.