Project Suncatcher: Google launches its first satellite to test whether its AI chips (TPUs) survive in space

Google is preparing to begin the orbital testing phase of Project Suncatcher, the "moonshot" project (very long-term and uncertain in outcome) investigating whether space could one day become a viable location for large-scale machine learning computing infrastructure. The announcement, signed by Travis Beals, Senior Director of Paradigms of Intelligence at Google, explains that the company introduced Suncatcher last year and that now, after years of preliminary research, it is set to launch a prototype satellite to assess how its TPUs (Tensor Processing Units, the AI chips designed by Google) behave under real spaceflight conditions.

The starting rationale is energy-related: in low Earth orbit (LEO), satellites can receive near-constant sunlight, which — according to Google's figures — would allow up to eight times more solar power to be generated than on the Earth's surface. The project's long-term vision is that, over time, it will be possible to link multiple satellite constellations together to handle large-scale AI workloads directly in orbit. Google compares this process to earlier "moonshots" such as autonomous driving or quantum computing, which required years of experimentation before becoming practical systems, and frames this first mission as a "measured and deliberate" step within that logic of working backwards from an ambitious goal.

This first test mission is flying as a rideshare payload on SpaceX's Transporter-18 vehicle, and has been developed in collaboration with Planet, the Earth observation satellite company. Its goal is to gather in-orbit data on how the TPUs withstand the physical stress of launch and the radiation and temperature extremes of space. Google notes that this initial launch will serve to identify failure points and apply those lessons to future missions, with a next milestone planned for 2027.

The first challenge is the physical survival of the hardware. A rocket trip to low orbit lasts around 10 minutes, during which the vehicle endures intense vibration and sustained acceleration loads of up to 10 times the force of gravity (10g), while individual components, such as the TPU chips themselves, can experience forces of between 50 and 100g. To simulate those conditions, the team subjected the satellite to vibration testing along its three axes, reproducing the frequencies typical of a real launch. According to the team itself, this kind of testing rarely goes exactly as planned, so they say they were "pleasantly surprised" that the hardware withstood the forces applied.

The second challenge is radiation. Outside the Earth's atmosphere, solar events and cosmic rays can seriously damage electronic components. To get ahead of that problem, Google tested its TPUs at a proton beam facility at the University of California, Davis' Crocker Nuclear Laboratory, running AI workloads during irradiation and monitoring how errors such as "bitflips" (accidental changes to a bit of information) affected processing. Initial results, according to the company, show that its Trillium TPUs perform "remarkably well" and can withstand a total ionizing radiation dose greater than they would receive during a five-year space mission. Even so, the team acknowledges that some aspects can only be verified in actual space, and that this first launch — scheduled, according to the text, for "next week" relative to the publication date — will provide the data needed to inform future missions.

The third challenge is cooling. TPUs generate a large amount of heat concentrated in a small area, but in the vacuum of space there is no air circulation, so heat can only be dissipated through radiators, which requires a completely different approach to cooling the electronics. Google is exploring several solutions, among them a combination of heat pipes and radiators, and has already tested this technology in a thermal vacuum chamber that simultaneously simulates the thermal and vacuum conditions of space. The team notes that it will refine these designs as it obtains more data from the orbital mission itself.

The fourth challenge, planned for a later phase of the project, is interconnectivity between satellites. The future design envisions satellites each carrying dozens of TPU chips, operating in orbital clusters, for which each satellite will need to know its own position and that of its neighbors with great precision, communicating via lasers. Google notes that the laser technology itself already exists, but that today's most advanced systems are optimized for low bandwidth over long distances, whereas this use case requires precisely the opposite: very high bandwidth over very short distances. Maintaining that link, they say, demands a precision they compare to "hitting a target the size of a coin from kilometers away while both points are in motion." This satellite-to-satellite laser link capability will be tested in 2027, when Google expects to put two such satellites into orbit.

It is worth noting that everything described comes entirely from Google's own corporate blog, with results and assessments ("pleasantly surprised," "perform remarkably well") reported by the company itself without any independent verification cited in the text. The article offers no investment figures, timelines for profitability or details on the program's cost, and devotes the bulk of its content to explaining the engineering challenges (vibration, radiation, cooling and laser communication) rather than announcing definitive results: Google itself frames this launch as an exploratory experiment within a multi-year research program, with no guarantee that AI computing in space will ever be viable at scale.