Google is preparing its first orbital hardware test for Project Suncatcher, a research effort focused on space-based artificial intelligence computing. Announced in November 2025, the project explores satellites carrying Google’s Tensor Processing Units, commonly called TPUs. Google plans to launch two prototype satellites with Earth-imaging company Planet during an early 2027 learning mission. The mission will test whether specialized AI hardware can operate reliably within a connected orbital system.
This flight will be small, but its purpose reaches beyond a routine component demonstration. Google wants to examine whether orbital solar power and optical networking could support large computing clusters. The concept remains experimental, and Google has not approved a commercial orbital data center. Results from the prototypes will guide later engineering choices rather than confirm the project’s ultimate feasibility.
How Project Suncatcher Would Work
Project Suncatcher proposes groups of solar-powered satellites that process machine-learning workloads using TPUs designed by Google. Those processors accelerate mathematical operations used for training and running many modern AI models. Free-space optical links would connect the spacecraft, allowing processors on separate satellites to exchange data. Ground stations would still handle commands, software delivery, results, and communications beyond the constellation.
The broader vision resembles a distributed data center assembled across many spacecraft. However, conventional data centers connect processors through short cables and carefully controlled networking equipment. An orbital cluster must send comparable information through lasers moving between fast-traveling satellites. That requirement makes communications and formation control central parts of the research.
The First Learning Mission
Google’s planned mission with Planet will place two prototypes in low Earth orbit. Planet brings experience building, launching, and operating compact commercial satellites. The spacecraft will carry Google computing hardware and systems needed to evaluate orbital performance. Google expects the experiment to provide practical evidence that ground testing cannot fully reproduce.
Engineers will monitor processor behavior, power consumption, thermal conditions, data errors, and communications. They must also study how hardware responds to repeated temperature changes and radiation exposure. The two-satellite mission cannot demonstrate a complete orbital computing center. Instead, it will test several assumptions supporting larger designs.
Testing Google’s AI Chips
Google has already exposed its Trillium-generation TPUs to proton radiation during ground experiments. The processors survived a cumulative dose exceeding the project’s estimated five-year orbital exposure, according to Google. Those results encouraged the company, but they do not cover every space hazard. Radiation can cause temporary errors, permanent damage, or corrupted information within processors and memory.
High-bandwidth memory appeared more sensitive than the TPU processors during Google’s tests. Engineers can address some errors through shielding, redundancy, software checks, and replacement capacity. However, each protective measure increases mass, cost, complexity, or electricity use. The orbital mission will reveal how the combined hardware behaves within an actual spacecraft.
Why Google Is Looking Toward Space
AI development requires growing amounts of electricity for processors, cooling systems, storage, and networking. New terrestrial data centers can also require land, water, electrical substations, and expanded transmission infrastructure. Google believes space could eventually provide another location for energy-intensive computing. That possibility depends largely on strong solar generation and falling launch prices.
Google’s analysis estimated orbital solar panels could produce up to eight times more energy than comparable terrestrial panels. Space-based panels avoid clouds, weather, and atmospheric filtering. A suitable sun-synchronous orbit could also provide extended sunlight while limiting eclipse periods. Nevertheless, satellites need batteries or operational adjustments whenever Earth blocks the Sun.
Power and Cooling Remain Difficult
Abundant sunlight does not automatically create a practical power system. Large solar arrays add mass and must survive launch vibration, deployment, radiation, and small orbital debris. Power electronics must convert and distribute electricity without creating unacceptable losses. Designers also need backup capacity when panels degrade or temporarily stop producing energy.
Cooling presents another major challenge because space contains no air for convection. Satellites must move processor heat through internal structures before radiating it outward. Powerful TPUs generate substantial heat while operating continuously. Large radiators could manage that heat, although they would increase spacecraft size, mass, and control requirements.
Building a Data-Center Network in Orbit
Modern AI training divides calculations across many processors that exchange information rapidly. Google says orbital links may need capacities reaching tens of terabits per second. Traditional radio connections would struggle to deliver those speeds between numerous satellites. Project Suncatcher therefore relies on laser-based optical communications.
Google reported laboratory optical tests reaching terabit-scale throughput across a one-kilometer path. Space removes atmospheric interference between satellites, but precise alignment remains essential. Narrow laser beams must track receivers while both spacecraft travel several kilometers each second. Vibrations, navigation errors, and changing positions could interrupt those connections.
Keeping Satellites Together
High-speed optical links work best when satellites remain relatively close. One Project Suncatcher design study modeled a cluster containing 81 spacecraft. Google’s orbital simulations suggested compact formations could remain manageable with carefully selected trajectories. Even so, operators would need accurate navigation, collision avoidance, and coordinated propulsion.
Every maneuver consumes propellant or shortens the life of another propulsion system. Spacecraft failures could also create collision hazards for the remaining cluster. Designers must ensure disabled satellites can leave orbit safely. These operational demands become more complicated as a formation gains additional processors and solar arrays.
Economics Could Determine the Outcome
Launching computing hardware remains far more expensive than transporting equipment to a terrestrial facility. Google’s study estimated that prices below $200 per kilogram could make orbital computing economically competitive. Some industry forecasts suggest that threshold might arrive during the mid-2030s. However, launch schedules, vehicle reliability, insurance, and spacecraft production would affect the complete calculation.
Comparisons must also include maintenance and equipment replacement. Technicians can repair terrestrial servers, replace failed cables, and upgrade cooling systems. Orbital hardware usually requires redundancy or another launch when important components fail. Rapid improvements in AI processors could make older satellites inefficient before their planned retirement dates.
Environmental and Regulatory Questions
Google presents Project Suncatcher as a possible response to AI’s increasing demand for energy and terrestrial resources. Orbital computing could reduce some land requirements and avoid water-based cooling. Yet launches produce emissions, while manufacturing satellites consumes materials and energy. Reentering hardware may also affect the upper atmosphere, although researchers continue studying those effects.
Large satellite clusters create additional concerns involving debris, astronomy, radio coordination, and crowded orbital regions. Operators would need launch approvals, communications licenses, and effective end-of-life plans. International coordination could become necessary as the system expands. Therefore, technical success would represent only one part of commercial readiness.
What the First Test Could Establish
The 2027 prototypes will not train frontier AI models at data-center scale. Their value will come from measuring real performance under orbital conditions. Engineers can compare those measurements against laboratory results and computer simulations. Unexpected failures could identify weak components before Google considers a larger demonstration.
If the mission succeeds, Google could test larger formations, faster optical links, or expanded solar systems. Poor results might lead engineers to redesign equipment or narrow the project’s goals. Either outcome would provide useful evidence about placing advanced AI infrastructure beyond Earth. For now, Project Suncatcher remains a research program rather than a promised replacement for terrestrial data centers.
