← All posts
This post is written in English.
Google's Project Suncatcher explores powering AI compute in space. Its prototype satellite launched on 1 October, and after weeks of commissioning, the team reached a "Hello World" milestone: "the first-ever AI response from a data center class TPU in orbit."
They uploaded a prompt during a ground-station pass, asking Gemma 4 for a haiku about being in space. The answer came back as the satellite passed Svalbard, Norway.
What it doesn't show
Google is clear about what's still ahead:
- Whether TPUs "can reliably execute workloads in space" is still to be verified.
- Thermal stress and radiation tests are planned next, with no results yet.
- Twin connected satellites are targeted for 2027.
- There's no throughput or long-duration performance data.
So it's a proof that the hardware switched on and ran something, which is a fair milestone for a first satellite.
Why anyone would try
On the ground, data centres are meeting local resistance, and power is a constraint. Solar power in space avoids some of that. It brings its own problems though: launch cost, radiation, getting rid of heat without air, and repairs that aren't really possible.
My read
I enjoy this story, and I'm sceptical of it at the same time. A haiku is a good demo and a tiny workload. The questions that decide this are economic: what does a useful amount of compute cost in orbit, compared with building on the ground, even with all the permits? Google calls this long-term research, and that seems like the right label. I'd wait for the 2027 satellites before reading much into it.
