Back to Inside Starcloud’s $2B Space Factory, with Philip Johnston & Ezra Feilden.
What Is Starcloud One Doing With Satellite Imagery?
Philip Johnston says Starcloud One runs inference on satellite imagery in orbit, so the coordinates of a vessel can come down in minutes rather than hours and days.
Starcloud One carries three chips from Nvidia, two from ARM, and CPUs from AMD. Earlier workloads included training NanoGPT from Andrej Karpathy, the first model trained in space, and running the first version of Gemini. The chips are mainly in a test phase now. The useful work is inference for government customers on synthetic aperture radar and hyperspectral data. Military satellites collect enormous amounts of imagery. Reaching a ground station takes at least an hour and a half, and the amount that gets downlinked is usually small. Optical terminals on the Starcloud satellite let that data arrive in real time. Inference runs on the satellite, and the result sent down can be the coordinates of a vessel.
The example is a government watching a patch of water to learn whether a vessel is crossing a boundary. The downlink is those coordinates, on a timeline of minutes instead of the current wait of hours and days.
Inside Starcloud’s $2B Space Factory ·
From the transcript
So you have five chips up in space right now, including one H100. Yes. What are they doing? Actually, we have three chips from NVIDIA, two from ARM, and we also have some CPUs from AMD on StarCloud One. What they're doing is, at the moment, we use them to do some interesting workloads like train the first model in space, NanoGPT from Andrej Karpathy.
We ran the first version of Gemini. But now they're actually doing useful workloads for government customers. So they are mainly in sort of as a test phase, but we're doing inference on satellite imagery. So on synthetic aperture radar data, on hyperspectral data. So we actually just... We'll do an announcement about some of the results of this, but we've got some very interesting results.