Google outlines ‘Project Suncatcher’, a space-based AI data centre plan

New Delhi: Google has unveiled “Project Suncatcher,” a project to scale AI compute in space using solar-powered satellite constellations equipped with Google TPUs and free-space optical links. 

Google stated that sun-synchronous, dawn-dusk low Earth orbits offer near-continuous sunlight, with solar panels capable of producing up to eight times more energy than on Earth. That reduces the need for heavy batteries and lowers the impact on terrestrial resources.

A modular design of smaller, tightly networked satellites is proposed instead of a single, monolithic craft. Early lab tests using commercial DWDM optics achieved 800 Gbps each way, or 1.6 Tbps total, across a short free-space path. The target is tens of terabits per second per inter-satellite link by flying in close formation.

Physics models suggest an illustrative 81-satellite cluster at about 650 km altitude could maintain next-nearest-neighbour distances of roughly 100–200 meters with modest station-keeping. 

Google also tested its v6e “Trillium” Cloud TPU under radiation. No hard failures were observed up to 15 krad total ionising dose on a single chip. The most sensitive elements were HBM subsystems, with irregularities beginning around 2 krad, which is still well above the shielded five-year mission estimate of 750 rad. 

On costs, Google’s analysis projects launch prices could fall below $200 per kg by the mid-2030s. At that level, operating a space-based AI data centre could approach the reported energy costs of terrestrial centres on a per-kilowatt-year basis. 

As a next step, Google plans a learning mission with Planet. Two prototype satellites are slated to launch by early 2027 to validate optical networking, models and TPU behaviour in orbit. 

The preprint, “Towards a future space-based, highly scalable AI infrastructure system design,” is authored by Blaise Agüera y Arcas, Travis Beals, Jessica V. Bloom and others at Google. The blog post is credited to Travis Beals, Senior Director, Paradigms of Intelligence.

Google cautions that major engineering challenges remain, including thermal management, high-bandwidth ground links and on-orbit reliability, but says early results show no fundamental blockers.