Google is preparing to launch a test satellite carrying Tensor Processing Units, or TPUs, to test whether AI hardware can operate reliably in space. The initiative, called Project Suncatcher, is a long-term research effort exploring the feasibility of placing machine learning infrastructure in low Earth orbit. Reuters reported that this is Google’s first orbital test for the project.
Why Google wants to put AI compute in space
The initial rationale is fairly straightforward: satellites in low Earth orbit can receive near-continuous solar energy, generating up to 8 times more solar power than on the ground, according to Google. If the concept works, multiple satellites could be networked into a cluster to process massive AI workloads rather than relying entirely on terrestrial data centers.
However, this is not yet a plan to build data centers in space. Google describes Project Suncatcher as a research moonshot. The first satellite is only intended to answer a fundamental question: can TPUs withstand the orbital environment? The data gathered will determine future launches.
The hardware will launch on SpaceX’s Transporter-18 rideshare mission, developed in collaboration with Planet. Google is targeting its next test milestone in 2027, when it plans to place two satellites into orbit to test communication links between them.
Challenges to overcome
The chip must survive three types of stress. The first is during launch. A flight to low Earth orbit lasts about 10 minutes, accompanied by intense vibration and continuous acceleration of up to 10 g. Individual components like TPUs can withstand forces of up to 50-100 g. Google stated that the team vibrated the satellite along three axes to simulate these conditions, and the hardware remained functional.
The second challenge is radiation. Outside the atmosphere, solar activity and cosmic rays can trigger electronic faults, such as bitflips. Google exposed a Trillium TPU to the UC Davis proton beam facility while running AI workloads. Preliminary results published by Google indicate that the chip withstood a total ionizing dose higher than what a 5-year space mission would be expected to endure. This remains a ground test, so data from orbit will be the true critical test.
The hardest part is likely thermal dissipation. TPUs generate significant heat in a confined space, yet in a vacuum there is no airflow for cooling. Google is testing a combination of heat pipes and radiators, which will then be evaluated in orbit to refine the design.
To process large AI workloads, satellites must communicate via laser
The long-term vision for Suncatcher will not rely on a single satellite. Google envisions each satellite carrying dozens of TPUs, flying in formation clusters and transmitting data to each other using lasers. Existing space laser systems typically prioritize low-bandwidth communications over vast distances. Google requires the opposite: very high bandwidth over short distances between co-moving objects.
The company compares the necessary precision to hitting a coin-sized target from miles away while both points are in motion. The 2027 inter-satellite link test will evaluate precisely this capability.
Project Suncatcher addresses a real-world pressure facing the AI industry: soaring power and cooling demands for data centers. Space offers abundant sunlight, but in exchange comes launch costs, latency, radiation, heat dissipation, and extremely challenging networking. This remains a research project, not a turnkey infrastructure solution. What will be worth watching following the first launch is real-world data on TPU durability and thermal management efficiency—two factors that will determine whether the dream of space-based AI compute is worth pursuing.