Google’s Project Suncatcher Satellite Runs AI in Orbit and Writes Its First Haiku

Google’s Project Suncatcher prototype, a satellite carrying one of the company’s AI chips, is online and has answered its first prompt from orbit, Google said Friday, Oct. 30. Engineers asked the onboard Gemma 4 model to write a haiku about being in space, in what Google called its “Hello World” milestone and the first AI response from a data-center-class TPU in orbit.

The satellite, built with Planet, launched Oct. 1 aboard SpaceX’s Transporter-18 rideshare mission from California. It is an early test of a long-term idea: that clusters of solar-powered satellites could one day run large AI workloads in space.

What the Suncatcher satellite did

In a blog post, Travis Beals, Google’s senior director for Paradigms of Intelligence, said the team spent several weeks on a commissioning checklist that included powering up the host and TPU trays and running early machine learning tests. During a ground-station pass, engineers uplinked the first prompt to Gemma 4, Google’s open model, and the poem came back as the satellite passed over Svalbard, Norway:

Blue marble below
Data streams in silent grace
Starlight feeds my core.

Google said it followed up with space-themed limericks and prompts asking the satellite to report the technical status of its payload. A spokesperson told TechCrunch the satellite was operating nominally.

The spacecraft carries a single Google Tensor Processing Unit and generates up to 1 kilowatt of electricity, TechCrunch reported. The chip’s power demands mean it can run for only about 20 minutes at a time, which complicates communication with the ground.

Testing radiation and heat

The next phase tests how the chips perform under thermal stress and space radiation. Radiation can cause “bit flips,” changing a stored value in computer memory and corrupting data without tripping typical monitoring tools. Ground testing in particle accelerators suggested the effect would be minimal for inference, the work of answering prompts, and a bigger problem for training models, TechCrunch reported.

“The error rate is very low if you’re thinking about typical inference operations, right? Like one in a million,” Beals told TechCrunch earlier this month. “On the other hand, it was already problematic for doing, say, some mega-scale training run where you’re going to have many thousands of chips running for months.” TechCrunch reported Google is confident its chips can handle inference workloads for a satellite’s five-year lifespan.

If the TPU holds up, Google plans to launch two more satellites in 2027 with higher power generation and optical links between them, a step toward the networked clusters the project envisions. Google called the twin interconnected satellites its next milestone. TechCrunch reported that Google’s concept for an orbital data center is a network of 81 satellites flying in close formation and processing in parallel. That sets it apart from rivals: TechCrunch reported that companies such as Starcloud and Cowboy Space plan to run inference on individual satellites, while SpaceX’s planned orbital data centers are expected to use laser links but fly too far apart to share processing effectively.

The data: the launch math

The biggest obstacle may be rockets. Suncatcher researchers published a peer-reviewed paper in the journal Joule examining how launch costs could fall. According to TechCrunch, they argue SpaceX has achieved a price-reducing “learning curve” of about 20% a year since Falcon 1, and that launch prices near $200 per kilogram by 2035 are plausible.

Reaching that price on the same trajectory would require Starship to fly about 370,000 tons of payload into orbit, or roughly 1,800 launches over 10 years, or 180 a year, assuming 200 metric tons per mission. Starship has never flown more than five times in a single year.

The two companies’ plans are linked: Alphabet, Google’s parent, holds a significant stake in SpaceX, TechCrunch noted. Satellites are also becoming targets, as shown by the KeyDrift hack against satellite firms.

Why it matters for AI’s power problem

Suncatcher arrives as AI spending strains tech balance sheets on the ground. This week Alphabet stock fell after its free cash flow turned negative on data center spending. Google has also been pushing new models such as Gemini 4 Carbon. Google describes Suncatcher as a research moonshot focused on infrastructure that could be economical in the 2030s, not a near-term product.

“Our chips have officially earned their space wings — an important achievement on the long road ahead for this cosmic moonshot,” Beals wrote.

Sources

• Google: We’ve received our first response from the Project Suncatcher prototype satellite
• TechCrunch: Google’s first data center satellite is online and writing poetry
• TechCrunch: Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers get off the ground
• Google: Our Project Suncatcher prototype satellite is in orbit