Google readies orbital launch of AI chips under Project Suncatcher

Google is preparing to launch a satellite carrying its Trillium TPU chips into low Earth orbit on a SpaceX Transporter rideshare mission, developed in partnership with Planet. The mission is the first hardware milestone of Project Suncatcher, a Google Research initiative exploring whether machine-learning compute can be run in space rather than on the ground.

The rationale is energy, not novelty: Google says a satellite in the right low-Earth orbit can receive up to eight times more usable solar power than an equivalent panel on Earth, which faces night, weather and atmospheric losses. Google reports that its hardware has already cleared vibration testing simulating launch forces of 50 to 100 g, and that the Trillium TPUs tolerate a total radiation dose greater than what they would accumulate over a five-year mission.

This first launch tests survivability, not a working data center: the satellite carries an undisclosed, small number of chips. The next milestone, planned for 2027, is a dual-satellite test of high-bandwidth laser interconnects — the piece of the puzzle that would let multiple orbital nodes function as a single cluster rather than isolated compute islands. Google says future satellite designs envision carrying dozens of TPU chips each.

Analyse Cardan-AI — For clients in energy and aerospace, the signal is less about space technology than about what it reveals: a hyperscaler with essentially unlimited access to capital and to terrestrial power purchase agreements is nonetheless testing whether it is cheaper to leave the grid altogether. That is a data point on how binding the energy constraint on AI scaling has become, and it changes how a site-selection or power-procurement decision for an AI data center should be benchmarked today.