Robots Atlas>ROBOTS ATLAS
Infrastructure

Project Suncatcher

2025ResearchPublished: 29 September 2026Updated: 29 September 2026Published
Key innovation
A Google research initiative to place TPU accelerators on low-Earth-orbit satellites and build scalable ML infrastructure powered by near-constant sunlight.
Category
Infrastructure
Abstraction level
System
Operation level
DeploymentSystem
Use cases
Scalable off-Earth ML training/inferenceExploiting near-constant solar powerRadiative cooling of AI workloadsDistributed on-orbit TPU clusters

How it works

A constellation of TPU-equipped satellites holds a tight formation, exchanging data over high-bandwidth optical links to form a distributed compute cluster. Power comes from solar panels (near-constant exposure); heat is rejected by heat pipes and radiators. Key challenges are radiation tolerance, precise formation keeping and inter-satellite link bandwidth.

Problem solved

Scaling terrestrial AI data centres runs into power, land and cooling limits. Suncatcher explores whether orbit — with abundant solar power and radiative cooling — could one day move part of ML workloads off Earth.

Components

Trillium-TPU satellitesCompute

Compute nodes of the constellation carrying Google TPU accelerators.

Optical inter-satellite linksNetworking

High-bandwidth laser communication binding the cluster.

Solar power and radiative coolingPower and thermal

Solar panels (up to 8× power) plus heat pipes/radiators.

Evolution

2025
Google announces Project Suncatcher
Inflection point
2026
Planned prototype launch (SpaceX Transporter-18)
2027
Planned two-satellite laser interconnectivity test