Earth's power and land can no longer keep up with AI's explosive appetite. The Wall Street Journal reports that Google is in early-stage talks with SpaceX to push data centers into orbit, turning the compute race into a space race.
Project Suncatcher: TPU Satellites by 2027
Google's classified project, Project Suncatcher, comes to light. In partnership with satellite imagery firm Planet Labs, Google plans to launch two prototype satellites carrying its own TPU AI chips by early 2027. If successful, the fleet would expand to 81 satellites flying in close formation. Benefits: solar power in space is 8x more efficient than on Earth; the vacuum environment provides natural radiation cooling, eliminating water-intensive cooling systems; and the project bypasses terrestrial power shortages, land acquisition battles, and NIMBY protests.
To ensure supply chain flexibility, Google is also talking to other rocket companies, but SpaceX's launch capability remains the top choice.
SpaceX's 1-Million-Satellite Network as IPO Showpiece
SpaceX itself is pursuing the same track. Earlier this year it filed with the FCC for permission to launch 1 million satellites, using continuous solar power in low Earth orbit to fuel AI computation. Musk calls it a "no-brainer" extension of the Starlink V3 platform. Wall Street analysts believe this orbital AI compute plan has become the most compelling selling point for SpaceX's upcoming IPO.
A strategic alliance would create perfect symbiosis: SpaceX provides unmatched launch capacity and Starlink global connectivity; Google brings cutting-edge TPU chips and cloud expertise.
Rivals Emerge: Starcloud Runs Gemini on H100 in Space
Competitors are piling in. Startup Starcloud already launched a satellite equipped with NVIDIA H100 chips and successfully ran Google's Gemini model in orbit. Both OpenAI and Jeff Bezos' Blue Origin have expressed strong interest.
Yet commercial reality remains daunting. Technical barriers include radiation-hardened chips to withstand cosmic rays, laser communication latency, orbit maintenance costs, and space debris collision risks. Launch costs: the entire industry's economics depend on SpaceX's Starship achieving large-scale reusability by the mid-2030s, cutting per-kg launch cost to tens of dollars. Regulatory uncertainty is also severe—SpaceX itself admitted in FCC filings that these are "unproven technologies."
With terrestrial energy hitting a ceiling, the migration of AI compute to space seems inevitable. But turning prototypes into a viable business will require overcoming a long list of technical and economic hurdles.