Google plans to fly its Tensor Processing Units (TPUs), the accelerators it uses to run AI workloads, on a prototype satellite aboard SpaceX’s Transporter-18 rideshare mission next week. The flight will gather data on how the chips handle launch stresses, radiation and temperature extremes in low Earth orbit, the company said Sept. 24.
The flight adds a hardware test before a separate two-satellite mission Google plans for 2027 to test laser links between spacecraft. Google first described the pair as a learning mission with Planet in 2025. Those links are central to its longer-term proposal because satellites would need to exchange data while orbiting close together; the upcoming flight does not test that network.
In its 2025 paper, Google said it tested a Trillium TPU with its host server under a 67-MeV proton beam intended to simulate radiation in a sun-synchronous low-Earth orbit. On one chip, an end-to-end machine-learning workload and compute-focused tests ran correctly through a cumulative radiation dose of 15 krad(Si), the maximum tested; separate high-bandwidth-memory stress tests showed irregularities after 2 krad(Si). For a modeled sun-synchronous orbit with substantial shielding, the paper estimated about 0.15 krad(Si) a year, or roughly 0.75 krad(Si) over five years. It also describes single-event faults: one energetic particle can trigger an instantaneous error such as a bit flip, and their effects on training jobs still need study.
The new flight is intended to add data from space. Google says its proposed heat-pipe-and-radiator cooling system has so far been tested in a thermal-vacuum chamber, because a vacuum has no airflow to carry heat away from the chips. The company has not yet reported how the hardware performs in orbit. The 2027 laser-link test remains a further milestone, not a demonstration of an orbital data center.
Sources
- Google Research, “Behind Project Suncatcher” (Sept. 24, 2026)
- Google Research, “Meet Project Suncatcher” (Nov. 4, 2025)
- Google, “Towards a future space-based, highly scalable AI infrastructure system design” (2025 preprint)
- Image: NASA Goddard Space Flight Center Scientific Visualization Studio, NASA’s orbiting Earth-observing fleet; illustrative only, not a Project Suncatcher image.