Running AI on Satellites Is Finally Leaving the Drawing Board

Big tech told Rama Afullo no. Both Google and SpaceX passed when he pitched the idea of running compute workloads directly in orbit. So he did what any stubborn builder should do: he quit, packed his bags, and started Satlyt. Now, fresh off an eight-million-dollar seed round, his company isn't even building hardware. They are skipping the rockets and focusing entirely on the stack. Smart move.

For decades, spacecraft have acted like glorified, expensive pack mules. They gather telemetry, snap blurry pictures, and wait for a slow, painful downlink to ground stations so human operators can fix anomalies. It is a wildly inefficient bottleneck. On a related note, satlyt is rewriting that script by deploying lightweight models – like Google DeepMind's Gemma – straight onto orbiting hardware. Slashing transmission sizes by over sixty percent isn't just a cool engineering flex; it saves hundreds of thousands of dollars per bird annually.

The hardware layer is finally — oddly—catching up, but the real moat will be the operating system. The If SpaceX and Google are building the space-faring equivalents of proprietary smartphones, Afullo wants to be the hardware-agnostic platform powering everything else. This week's launch tests this exact thesis, running NASA experiments and image-processing workloads on third-party spacecraft. They are acting as the VMware of the exosphere.

We love to see it. While the hype machine chases the next flavor-of-the-month LLM wrapper down here on Earth, real engineering is quietly happening four hundred miles above our heads. Building infrastructure that actually solves hard problems in hostile environments? That is the kind of craft we can always get behind.