Meet Project Suncatcher, Google’s plan to put AI data centers in space

But there is a physics problem. Received power decreases with the square of distance, so Google notes that satellites should maintain a proximity of one kilometer or less. This would require tighter training than any currently operational constellation, but it should be feasible. Google has developed analytical models suggesting that satellites hundreds of meters apart would require only “modest station-keeping maneuvers.”

Hardware designed for space is expensive and often less efficient than terrestrial systems, because the former must be hardened against extreme temperatures and radiation. Google’s approach to Project Suncatcher is to reuse components used on Earth, which might not be very robust when you put them in a satellite. However, innovations like the Snapdragon-powered Mars Ingenuity helicopter have shown that commercially available hardware can survive longer in space than we thought.

Google says Suncatcher only works if the TPUs can operate for at least five years, which equates to 750 rad. The company is testing this by projecting its latest Cloud TPU v6e (Trillium) into a 67 MeV proton beam. Google claims that while memory was most vulnerable to damage, experiments showed that TPUs can handle about three times as much radiation (almost 2 krad) before data corruption is detected.

Google hopes to launch two prototype TPU-equipped satellites by early 2027. It expects the cost of launching these first AI orbiters to be quite high. However, Google expects launch costs to drop in the mid-2030s to just $200 per kilogram. At this level, space data centers could become as economical as terrestrial versions.

The fact is that terrestrial data centers are dirty, noisy, and hungry for electricity and water. This has led many communities to oppose plans to build them close to where people live and work. Putting them in space could solve everyone’s problems (unless you’re an astronomer).