Google’s first prototype data centre satellite is online — and it is writing poetry. Project Suncatcher, the company’s research effort into space-based machine learning, has booted a TPU processor in orbit and returned a haiku generated by its Gemma model, in what Google describes as the first live test of AI compute beyond Earth. It is a fanciful milestone with serious intent, as our tech coverage explains.
Why Project Suncatcher Puts Computers in Space
The logic starts with energy. AI data centres on Earth are hungry for electricity and land, and their growth is colliding with grid limits. In orbit, solar panels receive far more sunlight than on the ground, unfiltered by atmosphere or night in the right orbits. Google’s long-term research asks whether clusters of satellites, linked by high-bandwidth optical connections, could one day train and run models using that abundant solar power.
Suncatcher is at the earliest, experimental end of that vision. The current prototype is a single small satellite proving out basics: that radiation-hardened or shielded processors can survive launch, operate in the space environment and communicate results back to Earth reliably.
A Haiku as a Proof of Life
The haiku returned from orbit — generated by a Gemma model running on the space-borne TPU — serves as the mission’s cheerful proof of life, demonstrating the full chain from power and compute to downlink. Engineers stress that the point is not the poem but the pipeline: inference running end-to-end on hardware designed for machine learning, in orbit.
As noted in TechCrunch’s report on the satellite, the project sits within a broader industry flurry of interest in orbital computing, with startups and space agencies alike exploring data centres above the atmosphere. Google’s entry, backed by its custom TPU silicon, gives the idea heavyweight credibility.
The Obstacles Are Considerable
Sceptics list formidable problems: launching compute is expensive, cooling hardware in a vacuum is hard, radiation degrades electronics, and maintaining or upgrading a satellite is nothing like swapping a server rack. Latency and bandwidth constraints also mean orbital computing suits some workloads far better than others.
Google frames Suncatcher as a moonshot research programme in the tradition of its X lab — the same culture that produced self-driving cars and stratospheric balloons. Most such projects never become products; the ones that do tend to change industries. For now, a small satellite writing haiku above the Earth is a beginning, not an answer.
Project Suncatcher’s orbiting haiku is equal parts stunt and signal. The economics of space-based AI remain unproven, but the first TPU has now computed in orbit — and the data centre map may never look quite the same.
