- Google plans to deploy AI data centers in orbit using solar-powered satellite constellations
- Each Suncatcher satellite would operate in a low sun-synchronous orbit, ensuring nearly continuous solar radiation.
- In laboratory tests, 1.6 terabits per second were achieved between two transceivers under controlled conditions.
Google’s “Project Suncatcher” reveals a bold idea: to install real data centers specialized in artificial intelligence in orbit around the Earth.
These orbital platforms would take the form of a constellation of compact satellites placed in a low sun-synchronous orbit between sunrise and sunset to capture nearly continuous sunlight.
Each device would contain machine learning hardware, including TPUs powered by solar energy captured more efficiently than on Earth’s surface.
A radical concept for orbital computation.
The purpose of this setup is to limit reliance on efficient energy storage systems while testing the feasibility of scalable and sustainable computing outside of Earth’s atmosphere.
The research team proposes communication between satellites at speeds comparable to terrestrial data centers.
With multi-channel wavelength multiplexing and spatial multiplexing, satellites can theoretically achieve transmission rates of tens of terabits per second.
To reduce the gap in signal performance, the satellites would fly within a few hundred meters of each other, enabling transfer rates already validated in laboratory tests of 1.6 Tbit/s.
Maintaining such dense formations requires sophisticated orbital control based on the Hill-Clohessy-Wiltshire equations, supplemented by careful numerical simulations to compensate for gravitational and atmospheric effects.
That’s why GooglingTests performed on the Trillium Cloud TPU v6e exposed to 67 MeV proton radiation revealed no critical damage, even at doses much higher than expected in orbit.
The most sensitive components, such as the high-bandwidth storage subsystems, showed only minor irregularities.
These results suggest that current TPU architectures, with minor modifications, could withstand low Earth orbits for longer missions.
However, the project’s financial viability is still uncertain. A significant reduction in implementation costs could make this solution viable.
If prices fall below $200 per kilogram by the mid-2030s, the cost of launching and operating data centers in orbit could approach the cost of terrestrial infrastructure, measured in kilowatt-hour equivalents per year.
However, this requires long-term reliability and minimal maintenance, two aspects that have not yet been tested on a large scale.
Despite encouraging signs, many parts of the Suncatcher project still rely more on theoretical models than empirical validation.
The upcoming partnership with Planet, which plans to launch two prototype satellites by 2027, will test optical interconnects and TPU performance in real-world orbital conditions.
The future of these orbital centers (simple research experiments or real pillars of infrastructure) will depend on further advances in energy management, communication stability and economic efficiency.