Google Research has run its Gemma 4 AI model on data-center TPU hardware aboard a prototype satellite in Low Earth Orbit, as part of its Project Suncatcher effort.
Google Research · Project Suncatcher
Gemma 4 runs on data-center TPUs aboard a prototype satellite in orbit
In its first "Hello World" test, the model was asked to write a haiku about being in space. The test was reported complete as the satellite passed over Svalbard, Norway.
How the first test ran
STEP 1
Power up
Team powers up data-center TPUs on the prototype satellite
→
STEP 2
Run model
Gemma 4 executes while the spacecraft is in Low Earth Orbit
→
STEP 3
Prompt
Ground engineers ask for a haiku about being in space
→
STEP 4
Test complete
Reported complete as the satellite passed over Svalbard
Flow of the first Hello World test as described in the announcement.
What the test shows
A modern large language model can be executed on accelerator hardware in orbit.
Computing in space is an alternative to ground facilities that need large amounts of electricity and cooling.
The demonstration is a concrete, if small, step for the Suncatcher effort.
What remains open
Radiation exposure, limited power, and servicing hardware after launch are not addressed in the announcement.
Performance, power consumption, test duration, and hardware configuration have not been disclosed.
No timeline has been given for next steps.
Bottom line
Read this as a proof of concept, not a commercial service.
Longer missions
Whether the prototype moves beyond a single test
Heavier workloads
Whether more demanding AI tasks can run in orbit
Operational use
Whether the idea reaches deployment, which remains undetermined
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...