Google launches its first space data center satellite which is now writing poetry in orbit

Google launched its prototype space data center earlier this month on a SpaceX Falcon 9 rocket, and according to TechCrunch, the spacecraft is now online, communicating with ground teams, and generating poetry in orbit. This early mission kicks off testing for future orbital networks designed to run large inference workloads away from Earth.

Key takeaways

• The prototype spacecraft carries four Google Tensor Processing Units and generates up to one kilowatt of electricity.
• Engineers tested the system by prompting the pre-loaded Google DeepMind Gemma 4 model to write a haiku.
• The current test campaign will check how the custom chips handle thermal stress and radiation in space.
• Alphabet holds an investment stake in SpaceX, and both firms rely on the Starship rocket for future launches.

What is a space data center and how does it work?

The new prototype from the Suncatcher project sits in orbit carrying four Google Tensor Processing Units. Power limits mean the hardware can only run a single processor for roughly twenty minutes at a time. After spending about three weeks testing onboard subsystems, ground engineers sent a prompt to the Gemma 4 model pre-loaded on the craft. The AI responded with a short poem about stars feeding its core and data streams flowing in silence.

TechCrunch reports that a company spokesperson confirmed the spacecraft operates nominally. Beyond creative writing, the team asked the hardware to report back on surrounding orbital conditions. This mission helps the company gather real data on how space radiation affects microchips outside of a terrestrial lab.

How do radiation and power limits affect Google tpu in space?

Particle accelerators help test processor durability on Earth, but engineers need to see actual conditions in orbit. High-energy radiation causes bit flips that alter stored values in computer memory and can corrupt data without tripping standard monitors. Ground simulations suggest radiation impacts inference workloads less than model training, but real flight data provides definitive answers.

The project plans to launch two more spacecraft equipped with advanced chips in 2027 if the current hardware proves capable. Those future units will generate higher power levels and test inter-satellite optical links. While competitors like Starcloud and Cowboy Space plan to run workloads on isolated units, Suncatcher aims for high-speed connections between spacecraft to handle collaborative workloads.

What happens next?

The ongoing flight campaign will measure the exact performance of the custom chips under thermal stress and radiation. Success could lead to the 2027 launches and longer-term plans for economical infrastructure in the 2030s. At the same time, the broader parent company faces financial pressures as integrating AI into search affects ad revenue growth, while future deployments depend heavily on Elon Musk completing a fully reusable heavy-lift Starship rocket.