Google’s AI Satellite Sends Its First Answer From Space: a Haiku

Share

Google’s Project Suncatcher satellite has sent back its first AI answer from space, and it was a poem. In a post on Google’s Keyword blog on October 30, Google said it sent a prompt up to the satellite, a Google TPU (the chip Google builds to run AI) booted up in orbit, and its Gemma 4 model wrote a haiku that came back down to Earth. Google calls it a “Hello World” moment: the first AI response from a data center class TPU in orbit.

Key Takeaways: Suncatcher’s first answer
First AI answer from orbit: Google ran a prompt on its Gemma 4 open model aboard the Suncatcher prototype satellite and got the full reply back on Earth.
It was a haiku: the first prompt asked for a short poem about being in space.
More tests done: Google also tried space-themed limericks and prompts that report the status of the satellite’s systems.
Next up: tests of how the TPUs cope with heat and radiation, then two linked satellites planned for 2027.
Nothing to download: this is a research project, not a product or feature you can use.

1. What happened up there

The launch: Google launched the prototype satellite on October 1. It spent the next few weeks working through its setup checklist, which included powering up the computer trays that hold the TPUs and running early machine learning tests.

The first prompt: during a pass over a ground station, Google sent a prompt to Gemma 4, the open AI model loaded on the satellite, asking it to write a haiku about being in space.

The reply: the answer came back as the satellite passed over Svalbard, Norway. Here it is, word for word:

Diagram of the Suncatcher test: prompt sent up, TPU boots, Gemma 4 writes a haiku, answer comes down over Svalbard
The four steps behind the first AI answer from orbit, with the haiku Gemma 4 wrote. Graphic: ChromeGeek. Source: Google

Why it matters: sending a prompt up, running it on an AI chip in space and getting the full answer back proves the basic setup works. It is a small job, but every bigger plan for Suncatcher depends on this step working first.

2. The satellite in numbers

DetailWhat we know
AI model on boardGemma 4, Google’s open model
Chips4 Google TPUs, per TechCrunch
PowerUp to 1 kW, enough to run one TPU for about 20 minutes at a time
LaunchedOctober 1, 2026
First answer receivedThis week, over Svalbard, Norway
Heat and radiation testsComing soon
Two linked satellitesPlanned for 2027

Power is the limit: because the TPU needs so much power, the satellite can only run one chip for short bursts, which makes timing tricky when commands and answers have to pass through a ground station.

Google Suncatcher team cheering around a laptop with the words Testing AI in orbit
Google’s Suncatcher team celebrating the mission. Image: Google

3. What Google tests next

Heat and radiation: Google says the next tests will check how its TPUs perform under heat stress and radiation in orbit. Radiation in space can flip bits in memory and quietly corrupt data, so this is the real test of whether the chips can be trusted with serious work.

Two satellites in 2027: if the chips hold up, Google’s next milestone is a pair of satellites linked together in 2027. The long-term idea behind Project Suncatcher is to power AI computing with solar energy in space.

Our take

A haiku sounds like a gimmick, but this is the AI version of a first phone call: it shows the whole chain works, from Earth to orbit and back. Nothing here changes your Chromebook today. Still, it is the same Gemma 4 family you can already run locally on a Chromebook, now running on a satellite, and that is a fun reminder of how far Google is pushing its AI chips.


Discover more from Chrome Geek

Subscribe to get the latest posts sent to your email.

Sophie is a Chicago-based editor for Chrome Geek covering Chromebooks, Gemini, and the Google ecosystem. When she is offline, you can usually find her watching Blackhawks hockey or grabbing iced coffee in the city.