Google's Project Suncatcher Tests TPUs in Orbit as a First Step Toward Space-Based AI Compute

Google Research disclosed details of Project Suncatcher, a moonshot effort to evaluate whether AI computing infrastructure — specifically its Tensor Processing Units (TPUs) — could operate in space.

Details

• First test flight: Trillium TPUs are flying aboard SpaceX’s Transporter-18 rideshare mission in partnership with Planet, exposed to launch forces up to 10g (with individual components experiencing 50-100g)
• Radiation resilience: Trillium TPUs have already demonstrated resistance to radiation doses exceeding what a five-year space mission would deliver
• Space-specific engineering: cooling relies on heat pipes and radiators built for vacuum conditions, and inter-satellite communication uses high-bandwidth laser links Google compares to “hitting a coin-size target from miles away while both points are in motion”
• Why space: low Earth orbit offers up to eight times more solar power than equivalent ground-based solar generation, motivating the eventual goal of satellite constellations that jointly handle substantial AI workloads
• Timeline: the initial prototype launch lands in late September 2026, with a two-satellite orbital test of the laser interconnect system planned for 2027

What happened next

Suncatcher is an early-stage research bet rather than a near-term product, testing the most basic assumption — that TPU hardware physically survives launch and space conditions — before Google commits to the harder problem of building and coordinating an orbital compute constellation.