Google’s “Project Suncatcher” paints a futuristic picture of AI in orbit, but the company’s own “cautionary note” highlights a terrifyingly practical problem: “on-orbit system reliability.” At 400 miles high, there is no IT support.
This is perhaps the “most significant” of the “significant engineering challenges” Google mentioned. On Earth, datacenters are built with redundancy, but they also rely on a constant stream of maintenance. A technician can simply swap a failed server, a dead fan, or a faulty cable.
In space, this is impossible. A single, unrecoverable error—a fried processor from a solar flare, a failed cooling component, or a broken optical link—would turn a multi-million-dollar AI satellite into a useless piece of space junk. The entire constellation must work flawlessly for its entire 5-to-10-year lifespan.
This challenge makes every other problem harder. The “thermal management” system must be 100% reliable. The “high-bandwidth” optical links must stay perfectly aligned. The TPUs must be shielded from radiation that can flip bits and corrupt data.
This is why Google is starting with just “two prototypes” in 2027. They aren’t just testing if the AI works in space; they’re testing if it can survive. Before this “moonshot” can become a reality, Google must master the art of building a computer that never breaks.