Google says its first Project Suncatcher satellite, built with Planet and launched on a SpaceX rocket, has demonstrated that AI computing can work in orbit, running a live inference task on a Tensor Processing Unit (TPU) in space.
Project Suncatcher · Google & Planet
First orbital AI test: a TPU in space runs a live inference task
Google says its first Project Suncatcher satellite, built with Planet and launched on a SpaceX rocket, showed that AI computing can work in orbit. A prompt sent from Earth was processed by a Tensor Processing Unit (TPU) in space, running inference on Gemma.
1
satellite in the first Suncatcher test
1
prompt round trip: Earth to orbit and back
Scale: orbital test vs. terrestrial data center
Accelerator count. The ground facility bar is cut off and is not drawn to scale.
Thousands
Terrestrial data center (cut off)
The gap is the point: one satellite with four chips proves the loop works, not that orbit can match ground capacity.
The proven loop
STEP 1
Request sent from Earth
→
STEP 2
TPU in orbit runs Gemma inference
→
STEP 3
Result returned for use
What the test supports
Basic physics of AI computing in orbit has been verified
A full Earth-to-orbit-to-Earth request cycle completed
Google describes it as a first step toward space data centers
What it does not show
Capacity or economics that rival ground facilities
Long-duration operation or durability
Cost of launch and satellite hardware versus ground builds
Open questions
The concept scales only if these factors are solved beyond a single experiment.
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...