Key Takeaways:
SpaceX's plan to deploy 100 gigawatts of AI compute in low Earth orbit by 2030 would require building the equivalent of 20 of Meta's planned Hyperion data centers — in space, on a timeline that begins mass production in less than 18 months.
Elon Musk's SpaceX on June 8 unveiled the AI1 satellite, a 150-kilowatt orbital compute module designed to bypass terrestrial power constraints that the company says will limit a $26.5 trillion AI total addressable market. The satellite, with a 70-meter wingspan and peak compute payload matching Nvidia's GB300 AI server rack at 150 kilowatts, will be mass-produced at a new 11-million-square-foot Gigasat factory in Bastrop, Texas — a facility more than 10 times the size of SpaceX's existing Starfactory complex.
"The AI satellite is much simpler than a Starlink satellite," Musk said in a 30-minute internal interview posted to X. "The AI satellite is essentially a lot of solar cells; you still need some laser links, but you don't have all of the super complex antennas. A lot of this is technology we've already manufactured for the Starlink V3 satellites."
The AI1 satellite delivers 120 kilowatts of sustained compute at a power density of 70 kilowatts per ton, drawing energy from a 150-kilowatt solar array built with SpaceX-manufactured cells. Heat dissipation — one of the hardest engineering problems in space-based computing — is handled by a 110-square-meter deployable liquid radiator with redundant pumping loops and integrated micrometeoroid shielding, achieving a thermal density of 1,400 watts per square meter. The compute module itself is interchangeable, meaning SpaceX is not locking the design to a single chipmaker.
The announcement comes days before SpaceX's anticipated IPO, expected to price on June 11 and begin trading on June 12, with a target raise of $75 billion and a valuation of approximately $1.75 trillion. In its IPO filing, SpaceX identified Earth's inability to rapidly expand power generation as the binding constraint on AI growth, positioning orbital data centers as the solution. The company's official filing targets 2028 as the start of commercial operations, though Musk's internal timeline is more aggressive.
SpaceX's production roadmap calls for an annual deployment rate of 1 gigawatt of orbital AI compute by the end of 2027, scaling to 10 gigawatts by 2029 and 100 gigawatts by 2030. Achieving the 2027 target would require launching more than 6,000 AI1 satellites in a single year — roughly 60 percent of the approximately 10,500 Starlink satellites currently in orbit as of June 2026.
The Gigasat factory, spanning 1,000 acres, will vertically integrate the entire AI1 supply chain under one roof: solar ingot and wafer production, solar cell manufacturing, printed circuit board assembly, silicon-based electronics, user terminals, ground gateways, and final satellite assembly. Solar manufacturing facilities are already under construction, with AI satellite production buildings set to break ground soon. Musk said the factory should reach "meaningful" mass production volumes by late 2027.
To put the 100-gigawatt target in perspective: Meta's Hyperion data center in Louisiana, the largest announced AI facility to date, is designed for a maximum of 5 gigawatts at a cost exceeding $100 billion, with its first 2-gigawatt phase not expected until 2030. xAI's Colossus 2 in Memphis, recently expanded to nearly 2 gigawatts with 555,000 GPUs at a cost of about $18 billion, is currently the world's largest single-site AI installation. One hundred gigawatts is equivalent to building 20 Hyperions or 50 Colossus 2-class facilities — in orbit.
SpaceX's cost advantage hinges on two factors: Starship's ability to dramatically lower per-kilogram launch costs, and a separate initiative called Terafab — a planned 1-million-square-foot factory where SpaceX, Tesla, and Intel would collaborate to produce custom AI chips at the 2-nanometer node, targeting annual output of 100 million to 200 million advanced chips representing 1 terawatt of compute. All three partners lack chip manufacturing experience at scale, and the 2-nanometer process is among the most technically demanding in the industry.
The economic case faces skepticism from established players. Blue Origin and Amazon founder Jeff Bezos, along with researchers including Andrew McCalip, have argued that expensive AI silicon combined with launch costs makes the current orbital data center model uneconomical. SpaceX has not disclosed per-satellite manufacturing costs or per-kilogram launch pricing for the AI1 program.
The AI1 satellites will operate in low Earth orbit at altitudes of 600 to 800 kilometers, with a one-way network latency of approximately 3 milliseconds — comparable to terrestrial fiber connections. Each satellite integrates laser inter-satellite links with up to 1 terabit per second of bandwidth, using Starlink's existing Ka-band and Ku-band antenna network or space-to-ground laser links for data downlink.
For investors, the stakes are binary. If SpaceX achieves even a fraction of its 2030 target, it could redirect billions in data center capital expenditure from terrestrial builders like Equinix and Digital Realty to space-based infrastructure. If the economics fail to pencil out, the Gigasat factory and AI1 program represent one of the largest capital allocation risks in the coming IPO. SpaceX shares, expected to begin trading at a $1.75 trillion valuation, already price in significant execution on the Starlink business; the AI compute thesis would represent a second, largely uncounted growth engine.
This article is for informational purposes only and does not constitute investment advice.
