Google to Launch Project Suncatcher Satellite With Trillium TPUs on October 1

Google is taking AI compute to orbit. On October 1, aboard a SpaceX Falcon 9, it will launch the first test of Project Suncatcher—a refrigerator-sized satellite carrying four Trillium TPUs powered by roughly one kilowatt of solar energy. The MVP will process short Gemini queries during 15-minute operating windows before it must pause for thermal cooldown, a candid acknowledgment of the brutal engineering constraints of computing in space.

The concept targets a looming bottleneck: terrestrial AI data centers are increasingly limited by power availability and cooling. Orbit offers uninterrupted solar power and the vacuum of space as a heat sink—at least in theory. The 15-minute windows and cooldown cycles show how far the technology is from practicality, but the test is about validating that TPUs can survive launch, radiation, and thermal cycling while doing useful inference.

Phase two, slated for 2027, will launch two satellites to test high-bandwidth laser inter-satellite links—the real prize, since networking distributed orbital compute is arguably harder than powering a single node.

The launch fits the week's compute-and-power theme from a radically different angle than xAI's terrestrial 1.44M-GPU cluster: where Musk is throwing gigawatts at the ground, Google is experimenting with escaping the grid entirely. Coverage framed it as Google entering 'the orbital AI race,' with skeptics questioning the economics—launch costs, latency to orbit, and maintenance make near-term commercial viability doubtful. But as a long-horizon bet on the power wall, and a signal of Google's full-stack ambition alongside Gemini 4, Suncatcher is a striking statement.