Google’s First Orbital AI Test Must Shut Down Every 20 Minutes
Google’s Project Suncatcher has reached orbit, but its first space-based artificial-intelligence computing test cannot run continuously. A company spokesperson said the prototype must power down every 20 minutes to prevent overheating, making thermal control the clearest dividing line between this chip experiment and a practical orbital data center.
The satellite launched Oct. 1 aboard a SpaceX Falcon 9 on the Transporter-18 rideshare mission from Vandenberg Space Force Base in California. Google describes Project Suncatcher as a long-term research effort examining whether machine-learning infrastructure could eventually operate in space. This first vehicle is meant to test how Google’s tensor processing units handle launch forces, orbital radiation and operation beyond the laboratory.
The shutdown is a thermal limit, not the whole mission
The 20-minute operating window does not make the satellite pointless. It can still collect data on processor behavior, radiation exposure and repeated operating cycles. Google said ground radiation tests at the University of California, Davis’ Crocker Nuclear Laboratory produced encouraging results, but some environmental conditions must be measured in orbit.
The shutdown requirement does, however, sharply limit what the spacecraft can prove. A computing platform intended to process workloads continuously must reach a steady thermal balance: heat generated by the processors has to move through the spacecraft and ultimately radiate into space at roughly the rate it is produced. This prototype is not designed to demonstrate that complete, continuous-duty system.
Space is cold, but vacuum is not an easy cooling medium. There is no surrounding air to carry heat away by convection. Inside a satellite, heat can move by conduction through boards, mounts, thermal straps and heat pipes. It then has to reach an external radiator, which rejects energy as infrared radiation toward space.
NASA’s small-spacecraft thermal-control guidance identifies limited surface area, tightly packed electronics and restricted volume as central problems for compact satellites. Solar cells, antennas and other hardware compete with radiators for exterior space. High-power processors intensify that conflict because dense electronics generate substantial heat in a small volume.
One kilowatt is far short of the projected operating scale
The New York Times reported that the prototype supplies about 1 kilowatt to its processors, although Google declined to confirm that figure to CNN. Even treated as a reported rather than company-confirmed number, it provides useful scale.
Google researchers have estimated that a data-center satellite could require approximately 50 to 100 kilowatts, comparable to a terrestrial data rack. Space-industry analyst Caleb Henry said current telecommunications satellites typically operate at about 10 to 20 kilowatts. On those comparisons, the test vehicle is not a small version of a finished orbital data center. It is a hardware experiment operating at a fraction of the projected power level.
That gap matters twice. More electrical power can support more computation, but power consumed by processors largely becomes waste heat. Scaling from roughly 1 kilowatt to tens of kilowatts therefore requires not just larger solar generation, but a correspondingly capable heat-transport and radiator system. Those additions consume spacecraft mass, volume and surface area and can complicate deployment and pointing.
Google plans to test radiators and heat pipes on the current vehicle, but chip performance remains the primary focus. Heat pipes can move energy from processors toward cooler spacecraft surfaces without a conventional mechanical pump. Radiators then provide the final path to space. The 20-minute shutdowns indicate that this prototype’s available thermal path cannot yet sustain its computing load indefinitely.
The 2027 satellites set a harder test
Google plans to deploy two additional test satellites in 2027 with partner Planet Labs. A company spokesperson said those spacecraft should operate continuously without overheating shutdowns and may also test laser communications between satellites.
Continuous operation would be a materially stronger milestone than simply running processors in orbit. It would show that power generation, heat transport, radiator capacity and workload scheduling can remain balanced over sustained periods. Testing a laser link would address another requirement: multiple computing spacecraft would need to exchange large amounts of data rather than behave as isolated processors.
Thermal performance is not the only long-term constraint. JLL Research noted that AI processors advance every one to two years, while satellites commonly remain in service for five to seven years. A computing spacecraft could therefore become technologically dated well before its platform reaches the end of its designed life. Launch cost, congestion and collision risk would add further constraints to any operational network.
For now, Project Suncatcher’s significance is narrower but concrete: Google has put its own AI hardware into orbit and can begin collecting flight data. The next decisive step comes in 2027, when the follow-on spacecraft are expected to attempt what the first one cannot keep computing without a forced shutdown every 20 minutes.
| More aerospace and engineering stories, right in your MSN feed. Follow AMI on MSN |
By David Whitaker — Associate editor for AMI’s aerospace and drone systems desk, translating flight systems, aircraft programs, spaceflight, and UAV developments into accessible technical stories.
