Google has launched a satellite to test Project Suncatcher, its plan to run artificial intelligence workloads directly in space using solar power.

The prototype, built by satellite maker Planet, carried four of Google’s Trillium TPU chips into orbit aboard a SpaceX Transporter-18 mission on October 1. The satellite generates about a kilowatt of solar power for short bursts of AI computation.

The idea is simple: data centres on Earth use huge amounts of electricity and water for cooling. In space, sunlight is abundant and the cold of space offers free cooling — though engineers say the biggest challenge is actually rejecting heat in orbit.

The test satellite will run 15-minute compute bursts so Google can measure performance and heat behaviour. The company said the experiment is an early research step, not a commercial service.

If the approach proves workable, orbital AI computing could one day reduce the energy footprint of machine learning. For now, the space-based TPU will face its first real test as engineers beam tasks up to it in the coming weeks.