QAZAQ GREEN.  Google has unveiled a bold research initiative known as Project Suncatcher, aimed at developing a scalable artificial intelligence (AI) infrastructure powered by solar energy in space. The concept envisions a constellation of satellites equipped with Google TPUs and interconnected through free-space optical links to perform large-scale machine learning computations beyond Earth.

According to Google’s senior director of Paradigms of Intelligence, Travis Beals, the project seeks to explore how AI computation could one day expand into orbit, where solar panels are up to eight times more efficient than on Earth and can generate power nearly continuously. “Space could become the next frontier for scaling AI compute,” he said.

The company’s new research paper, “Towards a future space-based, highly scalable AI infrastructure system design,” outlines the technical vision and early progress in addressing key challenges, including high-bandwidth communication between satellites, orbital dynamics, and radiation tolerance of computing hardware.

The proposed system consists of small, interconnected satellites operating in a sun-synchronous low Earth orbit to ensure constant sunlight exposure. This would allow the satellites to maintain continuous power generation while minimizing dependence on batteries. Google’s analysis shows that multi-channel optical transceivers and dense wavelength-division multiplexing could enable inter-satellite data transfer at speeds of tens of terabits per second.

Early laboratory tests achieved bidirectional transmission speeds of up to 1.6 Tbps, validating the feasibility of high-bandwidth optical links between satellites flying in close formation. Simulations also indicate that compact constellations, with satellites positioned just a few hundred meters apart, could maintain stable orbits with only minor corrections.

Radiation tests of Google’s Trillium v6e Cloud TPU showed strong resilience, with no critical failures recorded even under radiation doses far exceeding typical low-Earth orbit exposure levels.

Economically, Google’s analysis suggests that the falling cost of satellite launches — potentially below $200 per kilogram by the mid-2030s — could make space-based AI data centers cost-competitive with terrestrial facilities on a per-kilowatt-year basis.

To validate its models, Google plans a joint mission with Planet in 2027, involving two prototype satellites to test distributed AI computing and inter-satellite optical communication in orbit.

Project Suncatcher follows Google’s long tradition of ambitious scientific ventures, from early quantum computing research to the development of autonomous vehicles. If successful, the initiative could lay the groundwork for a new generation of space-based computing systems that combine renewable energy, advanced AI, and orbital engineering.