The future of artificial intelligence infrastructure may extend beyond Earth. Google has successfully brought its Project Suncatcher prototype satellite online, marking an important milestone in its research into space-based data centres.
Launched earlier this month in partnership with Planet Labs, the satellite carries four Google Tensor Processing Units (TPUs) and runs the Gemma 4 AI model. In its first successful test, engineers sent a prompt from Earth and received an AI-generated haiku from orbit โ demonstrating that the onboard computing system could process a request and communicate its response back to the ground.
Beyond this milestone, Google is evaluating how its AI chips perform under the extreme conditions of space, particularly radiation exposure, thermal stress, power limitations and data transmission. The prototype can currently operate only one TPU at a time, highlighting the engineering challenges that remain.
The next planned milestone is to launch two additional satellites in 2027 to test higher power generation and optical links between spacecraft. The long-term vision is to develop interconnected, solar-powered orbital computing infrastructure capable of supporting AI workloads.
Although commercial viability remains years away, this experiment could open an entirely new direction for data centre architecture, energy-efficient computing and the future of AI infrastructure.
