Google space satellite AI runs orbital haiku test for Suncatcher project

Google has recorded a milestone in its attempt to bring artificial intelligence to orbit: a satellite equipped with the company's custom AI chips has sent its first generated response from space. The test, part of the experimental Project Suncatcher, proved that Google's Tensor Processing Units (TPUs) can boot up and run an AI model outside Earth's atmosphere.

The satellite, launched earlier this month, powered on its TPUs and executed a prompt for the on-board Gemma 4 model to compose a haiku about being in space. Google said in a blog post that it also tried space-themed limericks and telemetry prompts, all of which the satellite handled successfully.

Project Suncatcher aims for orbital data centers

Project Suncatcher is Google's initiative to explore whether AI data centers could one day operate in space. The idea is to place energy-intensive AI workloads in orbit, where solar power is abundant and cooling is easier, potentially reducing the environmental footprint of terrestrial data centers. The successful satellite response is a preliminary step toward that vision.

Google's on-board TPUs — specialized hardware designed to accelerate machine learning tasks — have already proven effective in terrestrial data centers. The current test is focused on verifying that they can also function reliably in the harsh conditions of space, including radiation, temperature swings, and vacuum.

Next steps: stress tests and twin satellites

Google says the immediate next phase is a series of engineering evaluations to assess how the TPUs perform under thermal stress and radiation exposure in orbit. These checks will help the company understand whether its hardware can sustain long-duration workloads outside Earth's protective atmosphere.

If those tests go well, Google's roadmap includes launching twin interconnected satellites in 2027. This would allow data to be processed and shared between spacecraft, creating the foundation for a distributed space-based AI infrastructure.

The ability to run AI models in orbit could also enable real-time analysis of space-gathered data — such as Earth observation or satellite imagery — without having to transmit raw information to ground stations first. That would reduce latency and bandwidth requirements for future space missions.

Why space-based AI matters

For now, the satellite's haiku is a proof of concept, but the implications extend far beyond poetic output. Running AI models in space could pave the way for autonomous satellites, space-based research platforms, and even commercial AI services hosted from orbit. Google's Project Suncatcher is one of several efforts exploring this frontier, but the company's progress with TPUs marks a key technical hurdle being overcome.

The test also underscores how far AI hardware has come: a chip small enough to fit in a satellite can now execute a neural network that generates original text, all while hurtling through space. As Google continues its orbital evaluations, the prospect of AI data centers above the clouds is no longer science fiction — it's an engineering challenge with a timeline.

Google's next update is expected as the company wraps up the radiation and thermal stress tests and prepares for the twin-satellite milestone in 2027.