Google’s Project Suncatcher is a research effort to test machine-learning hardware in orbit. NPR reported on October 1, 2026, that the prototype’s four Tensor Processing Units (TPUs) ran a version of Gemma for simple queries for 15 minutes at a time because of heat-management constraints. The 15 minutes describes a reported computing interval; NPR separately reported a one-year mission goal. Those are different measures. NPR’s report
Why cooling is difficult in orbit
Computing hardware generates heat, and the vacuum of space has no surrounding air to carry it away through convection. Heat must be moved from the electronics to surfaces that can release it. Google describes research into heat pipes and radiators, along with tests of its cooling technology in a thermal-vacuum chamber. The company says the system’s performance in orbit remains to be learned. Google’s technical overview
NPR reported that radiators are among the heaviest components on the current mission, making heat rejection relevant to the spacecraft’s launch mass. Google’s approaches are part of ongoing research, not a demonstrated solution to the reported 15-minute interval.
The prototype and future designs have different configurations
The prototype uses a standard orbit and batteries. NPR corrected its earlier description to distinguish that configuration from a sun-synchronous orbit, designed to receive near-constant sunlight, in future satellite designs. The distinction matters: the reported interval belongs to the flown prototype, not a completed version of the future system.
Google confirmed that the prototype launched aboard SpaceX’s Transporter-18 and made contact. NeoTeo covered the initial contact in an earlier report; Google’s prototype update also described it as operating as expected.
Google’s proposed future satellites would carry dozens of TPUs and communicate over laser links. The company plans to put two satellites in orbit in 2027 to test those links, a step beyond the four-TPU prototype.
