Google's Satellite Is Writing Space Poetry—And It's Hilariously Bad

Forget launching rockets just to send people to space, Google is now sending AI into orbit to write bad poetry, and honestly, we're here for it.

According to TechCrunch, the search giant's Suncatcher prototype satellite, equipped with four Tensor Processing Units (TPUs), went live this week after launching earlier this month aboard a SpaceX Falcon 9. Engineers' first order of business: asking Google DeepMind's Gemma 4 model to compose a haiku about existence in space.

The result? "Blue marble below / Data streams in silent grace / Starlight feeds my core." It's not winning any Pulitzer Prizes, but there's something earnest about an AI waxing poetic about the cosmos from 300 miles up.

This isn't just a PR stunt, though the poetry angle is marketing. The satellite generates about 1 kilowatt of power, which means engineers can run only one TPU for roughly 20 minutes at a time. Every moment of computing, including that space haiku, is constrained by power.

The real test is radiation. TechCrunch reports that Google is using this mission to stress-test how its chips handle radiation in actual orbit. Particle accelerators on Earth can simulate cosmic radiation, but nothing replaces the real thing. Bit flips caused by radiation exposure can corrupt data in ways that normal monitoring systems might miss, especially during AI model training. This satellite is Google's early warning system for space-based computing.

The implications matter for the industry. If Suncatcher succeeds, Google plans to launch two more advanced satellites in 2027 with faster inter-satellite optical links, creating a distributed computing network in the sky. SpaceX has similar ambitions, though their satellites are expected to orbit too far apart for real collaboration. Google's bet is on tight, coherent coordination between spacecraft.

Cloud computing is going stratospheric. Within years, AI inference requests might be processed by satellites zipping overhead rather than ground-based data centers. The energy efficiency and latency improvements are significant.

For now, we're waiting for Google to ask their orbital AI to write sonnets about love and loss. If the haiku was any indication, the results will be unintentionally hilarious.