Google to launch first Project Suncatcher satellite to test AI chips in space

Orbital test setup and hardware

Alphabet's Google will launch an experimental satellite into low Earth orbit on October 1, 2026, marking the first orbital test of its Project Suncatcher initiative. The satellite, named MVP, will lift off aboard a SpaceX Falcon 9 rocket from Vandenberg Space Force Base in California as part of the Transporter-18 rideshare mission. Built in partnership with satellite company Planet Labs, the refrigerator-sized spacecraft integrates four custom Google Tensor Processing Units (TPUs). Together, these four processors supply the computing capacity of a single server in a terrestrial data center. Solar panels on the satellite will deliver one kilowatt of power, approximately the energy required to run a household hair dryer.

Thermal management and space radiation

Operating artificial intelligence accelerators in orbit presents engineering hurdles that differ from ground facilities. Because the vacuum of space prevents traditional fan cooling, Google developed a thermal system using heat pipes, external radiators, and layered conductive sheets of copper and aluminum. Under this design, the TPUs will run computational workloads in cycles of approximately 15 minutes before powering off to cool down. The mission will also evaluate how the processors withstand cosmic radiation, which can trigger bit flips that switch binary code between ones and zeros. Google plans to resolve computational errors during orbit by rebooting the chips when disruptions occur.

Pre-flight validation and testing

Prior to flight clearance, technicians subjected the hardware to terrestrial vibration and radiation testing. In a San Francisco laboratory, engineers mounted the satellite onto test tables to simulate the stress of a ten-minute ascent, where g-forces reach between 50 and 100 times gravity. Radiologic trials took place at the University of California, Davis Crocker Nuclear Laboratory, where Trillium TPUs operated inside a proton beam. Google confirmed the hardware absorbed a total radiation dose higher than the exposure expected over a five-year orbital lifespan. Once deployed, the satellite will spend one year answering test queries from ground stations, remaining in orbit for up to six years before atmospheric friction destroys it during re-entry.

Economic barriers and future laser links

Google established Project Suncatcher in November 2025 to investigate whether solar availability in low Earth orbit, where panels collect up to eight times more energy than on Earth, can power computing clusters. Other firms have also tested orbital compute, including Nvidia-backed startup Starcloud, which placed an H100 chip in orbit in November 2025 to run the Gemma model. Google initially planned a 2027 debut but accelerated testing by fitting its chips into an existing Planet Labs bus. To connect future clusters, Google aims to launch two satellites equipped with high-bandwidth laser links in 2027. Google research indicates launch costs must decline to around $200 per kilogram to achieve cost parity with land-based facilities.

Project Suncatcher senior director of product management Travis Beals outlined the testing philosophy for the mission.

Exploring space as a viable location for scalable AI compute won't happen all at once. It takes methodical engineering, starting with proving our hardware can handle the physical and unpredictable realities of operating in orbit. This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions.

James Manyika, Google senior vice president for research, technology and society, also noted that commercial applications remain distant.

We don't expect, to be perfectly frank, that we'll have anything usefully operational in the next few years.