Google's Project Suncatcher satellite runs Gemma 4 in orbit and returns a haiku
Google's AI chip-equipped satellite has produced its first response from space. In a blog post on Friday, Google said it had successfully booted up the satellite's Tensor Processing Units (TPUs). That let the on-board Gemma 4 model follow a prompt to write a haiku about being in space. The Verge reproduced the result:
Blue marble below Data streams in silent grace Starlight feeds my core.
Google says it also tried other space-themed limericks and telemetry prompts. The Verge's report does not say what those produced.
The satellite is part of Project Suncatcher, which The Verge describes as an experimental initiative that could eventually put AI data centers in space. Google launched the satellite into orbit earlier this month.
What comes next, in Google's words: "Next, we will continue with a series of tests to evaluate how TPUs perform under thermal stress and radiation exposure in orbit." Google adds that verifying its TPUs can reliably execute workloads in space "paves the way for our next milestone of twin interconnected satellites in 2027."
So the milestone is narrow. The chips powered up, a model ran on them, and an answer came back. Nothing in the report measures how well the hardware performs in orbit; that is what the planned tests are for.
Key facts
- In a blog post on Friday, Google said it booted the TPUs on its Project Suncatcher satellite, and the on-board Gemma 4 model followed a prompt to write a haiku about being in space.
- Google says it also tried other space-themed limericks and telemetry prompts.
- Google launched the satellite into orbit earlier this month; Project Suncatcher is described as an experimental initiative that could eventually put AI data centers in space.
- Next are tests of how the TPUs perform under thermal stress and radiation exposure in orbit.
- Google names twin interconnected satellites in 2027 as its next milestone.
Why it matters
This is the first reported sign that Google's AI chips, running a Gemma 4 model, work on a satellite in orbit. Project Suncatcher is an experimental initiative that could eventually put AI data centers in space, and a model answering a prompt from orbit is an early step toward that idea. It is a first step only: the report gives no performance data.
Who it affects
For now, mainly Google and the Project Suncatcher team, whose next steps depend on the planned tests. Anyone following the idea of AI computing in space gets a first data point. The report names no customers, partners or users.
How to use it
There is nothing to use yet. This is an experimental test, not a product or a service, and the report gives no pricing, access or availability. The thing to watch is the series of tests on thermal stress and radiation exposure, and the twin interconnected satellites Google plans for 2027.
How solid is it
The claim rests on Google's own blog post, as relayed by The Verge. The evidence offered is a short haiku generated by the on-board Gemma 4 model; Google also says it ran other limericks and telemetry prompts. There are no independent checks and no measurements in the report.
Risks and caveats
Google itself says the key questions remain open: how its TPUs perform under thermal stress and radiation exposure in orbit. The report gives no timescale for those tests and reports no results. It gives no figures for the number of chips, power, latency or bandwidth, and no cost or commercial timeline for space data centers. A haiku shows the system runs, not that it runs reliably.
“Verifying our TPUs can reliably execute workloads in space paves the way for our next milestone of twin interconnected satellites in 2027.”
— Google, blog post, as quoted by The Verge