Google Will Send AI Chips Into Orbit to Test Whether They Can Work in Space

Google is preparing to send its AI chips into orbit for the first time. A Project Suncatcher prototype satellite is scheduled to fly on SpaceX’s upcoming Transporter-18 mission, where it will test how the chips cope with launch forces, radiation and temperature extremes. The flight is an early check on Google’s much bigger idea: putting scalable AI computing infrastructure in space.

Why take AI hardware off Earth?

The attraction is solar power. Google says satellites in low Earth orbit can receive near-constant sunlight and generate up to eight times more solar power than on Earth. Project Suncatcher, announced last year, asks whether that setting could eventually support AI workloads at scale. The power opportunity is a reason to investigate, not evidence that an orbital computing system works.

Before Google can connect chips across a network of satellites, it needs to know whether its Tensor Processing Units, or TPUs, can operate on a spacecraft. These are Google’s AI chips. The prototype flight, developed with satellite company Planet, is designed to gather data that ground testing cannot provide about the conditions the hardware actually faces in orbit.

The launch is part of the experiment

The rocket trip will expose the satellite to intense shaking and acceleration. Google says individual components, including TPUs, can experience forces of 50 to 100 times Earth’s gravity during launch. Its team shook the satellite along three axes to simulate that stress and says the hardware held up. That result addresses the trip upward; it cannot show how the chips will perform after arrival.

Radiation presents a different test. At a proton-beam facility, Google ran AI workloads on Trillium TPUs while exposing them to radiation and monitoring errors. The company says its initial results showed the chips could withstand a total radiation dose greater than they would receive during a five-year space mission. That is a company-reported lab result, not a finding from orbit.

Heat has nowhere to ride

Cooling may prove harder to solve than surviving the ride. TPUs pack heat into a small area, but a spacecraft cannot rely on airflow in a vacuum. Google is testing heat pipes and radiators to move that heat away from the chips. Its team has tried the approach in a chamber that simulates vacuum and temperature conditions; the satellite will show how the new cooling system behaves in space.

Those measurements matter beyond a single chip. If the cooling design needs changes, Google can use the flight data to refine later satellites. The company describes this first mission as a way to find what works and where the hardware fails, rather than a demonstration of a finished orbital data center.

A working chip is not yet a network

Google’s longer-term design calls for satellites carrying dozens of TPUs each, flying in clusters to handle larger AI workloads. The satellites would need to exchange data through laser links. Google says its planned links require very high bandwidth over short distances, while the satellites keep track of their positions relative to one another. The company plans to test that connection with two satellites in 2027.

That makes the milestones distinct. The upcoming flight can reveal whether Google’s chips and cooling system operate in orbit. It will not answer whether moving satellites can maintain the connections needed for larger workloads. For now, the immediate result to watch is what the prototype teaches Google about keeping AI hardware running in space.