Google Launches Trillium TPUs Into Orbit for Space AI

Google, in partnership with Planet, has launched its first Project Suncatcher prototype satellite, designated M1, into a dawn-dusk sun-synchronous low Earth orbit. Riding aboard a SpaceX Falcon 9 Transporter-18 rideshare from Vandenberg Space Force Base, the refrigerator-sized spacecraft carries four Trillium v6e TPUs, which represent a hardware configuration equivalent to a Google Cloud TPU v6e-4 slice. The satellite draws approximately 1 kW of solar power to run Gemini inference workloads in short, 15-minute bursts, followed by dedicated thermal cooldown periods to manage heat in the vacuum of space.

Before the launch, Google subjected the hardware to rigorous testing. Vibration trials exposed components to forces between 50 and 100 times Earth's gravity. Radiation testing at the Crocker Nuclear Laboratory at the University of California, Davis, used a 67 megaelectronvolt proton beam. High-bandwidth memory irregularities began at a cumulative dose of 2 krad(Si), but the TPUs avoided hard failures up to 15 krad(Si), surpassing the radiation dose expected for a five-year mission. For future multi-satellite communication, Google also demonstrated a bench-scale optical system achieving 800 gigabits per second per direction, totaling 1.6 terabits per second of aggregate bidirectional capacity.

Google and Planet plan to launch two linked prototype satellites by early 2027 to test inter-satellite laser interconnects and distributed workloads in orbit. For AI practitioners, this shift to orbital computing changes how software must be architected. Instead of relying on the stable nodes of terrestrial data centers, developers must design systems around thermal scheduling, model partitioning across constrained optical links, and fault-tolerant inference that can quarantine unreliable chips. This requires new checkpointing and dynamic routing protocols to handle intermittent connectivity and thermal pauses.