Google will test its TPUs in orbit on SpaceX's Transporter-18

Why it matters

Google is testing whether its AI accelerators and cooling hardware can work in orbit, an early technical hurdle for its proposed satellite-based computing system. The flight will not prove that orbital AI compute can compete economically with terrestrial data centers.

Sundar Pichai (@sundarpichai), Google's CEO, said on September 24th that Project Suncatcher will send a prototype carrying Google's Tensor Processing Units into orbit on SpaceX's Transporter-18 rideshare mission. Built in partnership with Planet, the spacecraft is meant to test whether the chips can operate through launch and in space, not to demonstrate a working orbital data center.

https://x.com/sundarpichai/status/2103209164072010051

Google's technical update says the first flight will measure how the TPUs withstand the vibration of launch and the radiation and thermal extremes of low Earth orbit. Ars Technica reported that the mission is scheduled to launch on October 1st. That timetable makes this an engineering test on the near-term launch calendar, rather than the two-satellite demonstration Google had originally described for early 2027.

Pichai's post distilled the project to a simple question: "Can our TPUs survive and operate in space?" Google's answer remains an experiment. The company's October-bound prototype will gather flight data; the longer-term ambition is a network of solar-powered satellites that can run machine-learning workloads in orbit.

Google first introduced Project Suncatcher on November 4th, 2025, describing it as a research effort to explore scalable machine-learning compute in space. The initial plan called for two prototype satellites with Planet by early 2027. The current test brings forward an initial hardware check, while Google's stated next milestone remains a 2027 flight of two satellites to test high-bandwidth laser links between spacecraft.

The test addresses several problems that do not show up in a terrestrial data center. Google says a rocket's vibration and acceleration put hardware under intense mechanical stress; once in orbit, radiation can cause errors in electronic components. The company's ground tests included shaking the satellite along three axes and running AI workloads on Trillium TPUs in a proton-beam facility at UC Davis's Crocker Nuclear Laboratory. Google says those chips withstood a radiation dose greater than the amount expected over a five-year mission. Those are company-reported test results. The orbit flight is meant to establish how the hardware performs in actual conditions.

Heat presents a separate constraint. In a vacuum, conventional airflow cannot carry heat away from processors. Google says its design combines heat pipes and radiators and has been tested in a thermal-vacuum chamber on Earth. Ars Technica reported that the first satellite carries four TPUs and can run them for only about 15 minutes at a time before its cooling system needs to catch up. That limit puts the mission's scope in focus: the prototype tests whether components and cooling concepts can work in orbit, rather than delivering continuous computing service.

The economic case depends on solving those physical problems at a much larger scale. Google argues that satellites in low Earth orbit can receive near-constant sunlight and says they could generate up to eight times more solar power than on Earth. But a solar supply advantage alone does not establish that orbiting compute can compete with terrestrial facilities. The satellite has to manage power, heat, communications, maintenance and launch costs as one system. Google has described laser links as essential to connecting future satellites, and says the 2027 two-spacecraft test will examine that technology.

For now, Project Suncatcher is a hardware-validation program with a high-profile destination. Pichai's announcement puts Google's custom AI chips on an upcoming SpaceX flight; the result that matters will be the operating data Google brings back, including how the chips behave under radiation and how long the cooling system can support workloads. Google's own update calls this first launch a way to identify failures and inform later missions. A successful flight would answer a narrow question about operating TPUs in orbit. It would leave the harder question - whether a connected fleet can provide useful, economical AI compute - for later tests.