Google Project Suncatcher: TPUs head to space for first in-orbit test

Google will send its own AI chips into orbit for the first time next week as part of Project Suncatcher, a prototype satellite built with Planet that carries Tensor Processing Units (TPUs) aboard SpaceX's Transporter-18 rideshare mission. The goal is to test whether TPUs can withstand launch G-forces, space radiation, and vacuum cooling, a critical step toward determining if orbital data centers could be feasible.

Why orbital AI compute matters for builders

If space-based AI computing becomes viable, it could change the economics of training and inference. Google notes that solar panels in low Earth orbit can collect up to eight times more power than on the ground. For AI workloads that are power-hungry, that's a significant advantage. But the engineering challenges are steep: launch stress, radiation, and cooling in a vacuum.

This test isn't about deploying production workloads yet. It's about validating whether the hardware can survive at all. For builders designing edge AI systems for harsh environments, the radiation and thermal management data from this mission could inform future rugged designs.

What the hardware test actually covers

Google has already put TPUs through ground-based stress tests. The team shook the satellite on all three axes to simulate launch forces of 50 to 100 g, and the hardware held up. For radiation, TPUs were exposed to a proton beam at UC Davis's Crocker Nuclear Laboratory and withstood a larger total dose than expected over a five-year mission, according to Google.

Cooling is another major hurdle. Without air, heat can only escape through radiators. Google is testing a combination of heat pipes and radiators, which has so far worked in vacuum chamber tests on Earth.

The next milestone is a 2027 mission where two satellites will test laser links for TPU clusters, enabling satellite-to-satellite AI networking.

The long road to space data centers

Even if the hardware passes in-orbit testing, economics remain the biggest obstacle. Google's own research estimates that launch costs would need to fall to about $200 per kilogram for space computing to compete on cost. Current rideshare pricing is higher, and while Starship may lower costs, that's not yet a given.

Google is not alone in this race. Nvidia-backed startup Starcloud launched an H100 chip into orbit last November and ran Google's Gemma model. SpaceX itself has plans for Nvidia-powered AI satellites.

For AI builders, the immediate takeaway is that orbital compute is still years away from being practical. But the engineering validation happening now will determine whether it becomes a realistic option. If you're building systems that need ultra-low latency from orbit or massive solar power, keep an eye on Project Suncatcher's results.