Project Suncatcher tests TPUs in space aboard SpaceX’s Transporter-18

Google is sending its AI brains to space. Project Suncatcher, a collaboration between Google and Planet Labs, will fly a prototype satellite equipped with four Tensor Processing Units aboard SpaceX’s Transporter-18 mission, scheduled for October 1, 2026. The goal: figure out whether low Earth orbit is a viable home for the kind of machine learning compute that’s currently straining terrestrial data centers to their limits.

The prototype is roughly the size of a refrigerator. It carries four Trillium-generation TPUs, Google’s custom-designed AI chips, which will need to survive the violent vibrations and g-forces of launch before facing an even less hospitable workplace: the radiation-soaked vacuum of space.

Why put AI chips in orbit

The logic is surprisingly straightforward. Satellites in dawn-dusk sun-synchronous orbits enjoy near-constant sunlight, potentially generating up to eight times more solar power than equivalent panels on the ground. For AI workloads that devour electricity like few other computing tasks, that’s an enormous advantage.

The satellite won’t be running marathon AI training sessions, though. Due to cooling limitations in the vacuum of space, the TPUs will operate in short bursts of roughly 15 minutes at a time. During those windows, the satellite will run Gemini models, Google’s flagship family of AI systems, to benchmark real-world performance in orbital conditions.

Surviving the harshness of space

Before committing hardware to a rocket, Google ran extensive ground-based radiation testing. The results were encouraging: TPUs, including their particularly sensitive high-bandwidth memory components, withstood radiation doses of 2-15 krad without experiencing hard failures. That threshold exceeds the radiation exposure expected over the satellite’s projected five-year mission lifetime.

The bigger constellation vision

One satellite is a proof of concept. Google’s longer-term ambition involves a constellation of approximately 81 satellites connected via high-bandwidth free-space optical communications, essentially laser links between spacecraft.

Google plans to test two-satellite laser link configurations in 2027, a critical step toward proving that distributed orbital compute can actually function as a coherent system.

Project Suncatcher was officially announced in November 2025. Google has been clear that Suncatcher is a research project, not an imminent commercial product. There’s no timeline for when orbital TPU clusters might handle production workloads. The Transporter-18 mission is about answering fundamental questions: Can these chips survive launch? Do they perform reliably in radiation environments over months and years? Can they be cooled effectively enough to do useful work?