Google plans solar-powered orbital AI data centers with TPUs

Google has unveiled an ambitious effort called Project Suncatcher—an initiative to place AI data centers on satellites in orbit, powered exclusively by solar energy. The goal is to tap an effectively boundless clean energy source to run the company’s powerful TPU processors for both training and inference.

On Earth, Google’s data centers already raise concerns over massive electricity use and pressure on power grids. In orbit, however, solar panels can operate almost continuously and more efficiently than their ground-based counterparts. Google engineer Travis Beals said that, looking ahead, space could prove the most suitable place to scale AI computing.

The project, though, has to clear serious engineering hurdles. One is radiation. Google is testing its Trillium TPUs to ensure they can withstand the extreme conditions of space. Early results indicate the processors can function for up to five years without degradation.

Another challenge is ultra-high-speed data exchange between orbital facilities. To tackle this, Google plans to use laser optical communication links capable of moving data at tens of terabits per second. The satellites would need to fly within a few kilometers of one another—an arrangement that raises the risk of collisions and damage from space debris, where a neat formation on paper still demands exacting control in practice.

Despite the price tag, Google views the plan as economically promising. Internal projections suggest that by the mid-2030s, operating costs for orbital data centers could match those of terrestrial ones, helped by cheaper rocket launches. The competitive field is already forming: SpaceX and Jeff Bezos’s Blue Origin are also exploring ways to put compute power into orbit. The race here is as much about positioning as it is about engineering.

For now, Google is at an early stage. In partnership with Planet Labs, the company plans to launch two prototype satellites by early 2027 to test hardware resilience and gauge the concept’s viability. If those trials go well, Project Suncatcher could open a new chapter for energy-sustainable AI computing—quite literally beyond Earth.