SpaceX announced an 11-million-square-foot manufacturing facility in Bastrop, Texas, called Gigasat, dedicated to producing orbital AI data center satellites.
The company is targeting 1 gigawatt of space-based AI compute capacity by the end of 2027, with plans to scale that figure by an order of magnitude annually thereafter, with Elon Musk expressing ambitions for 100 gigawatts per year by 2030.
Two prototype satellites are planned for launch in early 2027, with the business model mirroring SpaceX's terrestrial strategy: lease the compute.
The premise behind the project is that ground-based AI infrastructure is running into hard physical limits. Terrestrial grid constraints are projected to restrict roughly 40 percent of AI data center expansion by 2027, and the top five hyperscalers alone plan to spend $602 billion on infrastructure in 2026.
SpaceX's orbital approach is designed to bypass those limits by leveraging near-continuous solar energy in low Earth orbit, with waste heat radiated directly into the vacuum of space rather than managed through water-cooling or air systems.
The company already has existing compute revenue to point to: Anthropic is reported to be paying roughly $1.25 billion per month to rent xAI data center capacity, and Google has agreed to pay approximately $920 million per month for AI capacity.
The scale of the ambition, however, sits well ahead of what has been demonstrated. SpaceX's own pre-IPO S-1 filing acknowledges significant technical complexity and unproven technologies, including radiation hardening of chips in orbit and latency constraints that could limit which AI workloads are viable candidates for orbital deployment.
SpaceX is also not the only company pursuing this direction. Blue Origin has filed with the FCC for a constellation of 51,600 data center satellites, Google's Project Suncatcher targets orbital AI chip deployments as early as 2027, and China's Adaspace is pursuing similar concepts.
Whether orbital compute eventually competes with terrestrial infrastructure on cost is the question the Gigasat announcement does not answer.
What the project does reflect is a broader recognition among large infrastructure players that the power and land constraints shaping ground-based AI expansion are a real challenge, and that the investment case for alternatives is worth taking seriously even before the economics are fully resolved.
