Google is set to launch its MVP satellite, featuring AI chips, as part of Project Suncatcher, marking a significant step in space computing.
Google is preparing to launch its MVP satellite, which features custom AI chips, as part of its Project Suncatcher initiative. This marks the first orbital test of the project aimed at enhancing computational infrastructure for AI in space, according to reports.
The satellite will be sent into orbit on SpaceX's Transporter-18 mission, utilizing a Falcon 9 rocket from Vandenberg Space Force Base in California. Developed in partnership with Planet Labs, the MVP is approximately the size of a refrigerator and equipped with four Tensor Processing Units (TPUs), providing computational power on par with a single server in a data center. The satellite will be powered by solar panels generating around 1 kW.
Once in orbit, the MVP will process AI queries and evaluate the performance of Google's Gemini AI system. The satellite is expected to function for about one year, although it could remain in orbit for up to six years before re-entering the atmosphere.
Assessing Performance in Harsh Conditions
The mission's main goal is to test the resilience of Google's AI accelerators against radiation and vacuum conditions encountered in space. Engineers have conducted vibration tests to simulate the launch experience, and the TPUs have also undergone radiation testing at the Crocker Nuclear Laboratory, successfully enduring a radiation dose expected over five years in orbit.
Cooling poses a notable challenge, as heat dissipation in a vacuum differs from terrestrial environments. Google has developed a system utilizing conductive materials and radiators to manage heat, which has been validated in thermal vacuum tests. Currently, the TPUs can operate continuously for about 15 minutes before requiring cooling, which is adequate for Gemini tasks but not yet sufficient for continuous data center operations.
Future of Space Computing
Announced in November 2025, Project Suncatcher envisions a constellation of satellites with TPUs and solar panels linked through high-speed laser channels. Operating in low Earth orbit, these satellites could generate significantly more energy than on the ground, potentially reducing reliance on terrestrial power grids.
For effective network integration, rapid communication between satellites is essential. Google has achieved impressive data transmission rates of 1.6 Tbps under lab conditions. Plans for 2027 include launching two additional prototypes to further test interoperability and laser communications, with future models expected to carry dozens of TPUs each.
However, Google does not foresee a fully functional orbital AI infrastructure in the immediate future. They project that as launch costs decrease to about $200 per kilogram, space computing could become more economically viable compared to traditional data centers by the middle of the next decade.
