Google sent four of its AI chips into orbit aboard a refrigerator-sized satellite on 1 October 2026, NPR reported, beginning the spaceflight test described in the company’s Project Suncatcher plans. The Planet-built spacecraft is a prototype for Google’s proposal to put solar-powered computing infrastructure in space. Its immediate job is narrower: to find out how the chips cope with the journey and whether they can run once they arrive.
Key points
- The Planet satellite carries four Google Tensor Processing Units.
- Google plans to run its Gemma model in 15-minute bursts while the chips’ cooling system is tested.
- Google says ground tests exposed its Trillium chips to a radiation dose greater than that expected over a five-year space mission.
- Google plans a two-satellite test of laser communications in 2027.
Four TPUs, 15 minutes at a time
The satellite’s four Tensor Processing Units, or TPUs, are chips Google designed for machine learning and already uses in data centres on Earth. On this mission they will run a version of Google’s open-weight Gemma model to answer simple queries for 15 minutes at a time, NPR reported. Google plans to stop each run so the radiators carrying heat away from the chips can cool before another begins.
A TPU does the calculations a model needs to produce an answer. Putting one in orbit changes the conditions around that calculation: the chip must survive launch forces and radiation, receive power from the spacecraft and lose the heat it produces while working. Google says this mission is intended to gather data on those physical demands, rather than to operate a full data centre. Its earlier orbital chip test plan set out the 1 October launch date.
Answering a simple question could eventually involve a calculation made in orbit, if processing there could be sustained and the answer sent back to Earth. In this test, a question would have to fit within a short operating period before the chips pause for cooling. That makes the familiar act of getting an answer a useful distinction between what the prototype is designed to attempt and what Google hopes an orbital network could do.
Google’s satellite will travel in a sun-synchronous orbit, where its panels will almost never enter shade and it will not need heavy batteries or backup power. Google says satellites in low Earth orbit can generate up to eight times more solar power than panels on Earth. The attraction for a computing system is access to that sunlight, though running chips also creates heat that the spacecraft must remove.
Trillium chips meet radiation and vacuum
Before launch, Google shook the spacecraft along three axes to imitate rocket vibration. It also ran AI workloads on its Trillium TPUs while exposing them to a proton beam at UC Davis’s Crocker Nuclear Laboratory. Google says its initial results found that the chips could withstand a total ionising dose greater than they would receive during a five-year space mission. Those measurements came from ground tests; the orbiting prototype will put the hardware through the conditions of spaceflight itself.
Cooling calls for a different route out of the spacecraft. Air and water can carry heat away from computers on Earth, but a satellite cannot shed heat into the surrounding vacuum that way. The prototype places thermal interface material between the chips and aluminium and copper heat pipes, which carry heat towards radiators, Ars Technica reported. Think of the pipes as a path from a hot component to a surface that can release the heat into space.
Google says it tested its cooling approach in a thermal vacuum chamber before launch. Travis Beals, the senior director leading Project Suncatcher, said the radiators are among the heaviest parts of this mission, with greater mass adding to launch costs. The 15-minute operating periods put that constraint into the test itself: running the model is only part of the task when the system must also recover enough to run again.
Maintenance presents another physical obstacle. In a data centre on Earth, failed equipment can be reached and repaired; in orbit, that access is much harder, said Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University. Spacecraft also face radiation that can interfere with calculations. Google’s proton-beam work monitored errors during AI workloads, while the satellite test is intended to supply data from operation beyond the atmosphere.
Planet’s satellite precedes Google’s laser test
Project Suncatcher reaches beyond a single spacecraft. Google envisages clusters whose satellites each carry dozens of TPUs and exchange information by laser. Those links would let separate spacecraft take part in larger AI workloads, but Google says they would require high bandwidth over short distances while the satellites continually move relative to one another. The company first revealed the project in November 2025 as an exploration of scalable machine-learning infrastructure in space.
The distance between this prototype and that arrangement is substantial. Large AI data centres being built on Earth can contain hundreds of thousands of chips, Scientific American reported, while the Planet satellite carries four. Alan George, a professor of electrical and computer engineering at the University of Pittsburgh, also identified transmission back to Earth as a difficulty for space-based computing. An orbital system would have to move its answers home as well as distribute work among its satellites.
Planet will bring the newly launched spacecraft into operation before Google starts testing the TPUs, and Beals told NPR the mission’s goal is to remain operational for a year. Google plans to put two satellites in orbit in 2027 to test the laser connections intended for later clusters.
