Imagine data centers floating in orbit, powered directly by the sun, running machine learning models beyond Earth’s energy constraints.
It’s Project Suncatcher, Google’s bold new moonshot. This is a research initiative exploring how solar-powered satellite constellations could one day scale artificial intelligence (AI) computing in space.
The project, announced this week, envisions a compact satellite network powered by Google’s Tensor Processing Units (TPUs), interconnected through optical links and powered by near-continuous sunlight.
The goal is to build a space-based AI infrastructure that has the potential to revolutionize the way machine learning is trained and deployed.
“The sun is the ultimate energy source in the solar system, emitting more than 100 trillion times more power than humanity produces,” Google said in a preprint paper titled “Towards future space-based, highly scalable AI infrastructure system design.”
“With the right orbit, solar panels can be up to eight times more productive than on Earth and generate electricity almost continuously.”
The study outlines an ambitious system that combines solar power generation, low-Earth orbit positioning, and optical networks to create an interconnected “AI constellation.”
Google said this approach “minimizes the impact on ground resources while freeing up unprecedented computing power for AI workloads.”
Solar power, infinite computing
The proposed system would consist of a satellite flying in a sun-synchronous orbit, allowing near-constant sunlight capture while reducing reliance on batteries.
Each satellite carries computing hardware, specifically TPUs, connected by free-space optical links that can transfer data at tens of terabits per second.
Large-scale machine learning tasks require huge amounts of bandwidth, which is typically achieved through vast ground-based data centers.
To recreate that in space, Google researchers are proposing densely formed satellite clusters in which spacecraft fly just a few hundred meters apart. This proximity helps overcome signal power losses that typically limit long-range space communications.
“Our analysis shows that this is possible with multichannel dense wavelength division multiplexing (DWDM) transceivers and spatial multiplexing,” the team writes. Early tests have already achieved 800 Gbps each way between prototype systems, suggesting that multi-terabit interlinks are possible.
Radiation resistance is also an important issue. The company tested its Trillium TPU v6e chip under a 67 MeV proton beam and found it to be surprisingly resistant. “Up to the highest tested dose of 15 krad(Si), no hard failures due to TID occurred,” the paper states.
orbit the future
Project Suncatcher is still in the early stages of research, but Google plans to test the concept in space soon.
The company, in collaboration with Planet Labs, aims to launch two prototype satellites by early 2027. These will test the TPU hardware and validate optical intersatellite communications in orbit.
Economic feasibility also appears to be within reach. Satellite launch costs continue to fall, with prices projected to fall below $200 per kilogram by the mid-2030s, and Google’s analysis shows that the economics of space-based data centers could be on par with Earth-based systems.
The project joins a long line of moonshots for Google, from quantum computing to self-driving cars.
“As with all moonshots, there will be some unknowns, but in this spirit we are setting out to build technology that once seemed unrealistic,” the team said.
With Project Suncatcher, Google is moving AI beyond the cloud and onto a trajectory where the next computing revolution may be powered by solar power.
With over a decade of career in journalism, Neetika Walter has worked with Economic Times, ANI and Hindustan Times covering politics, business, technology and clean energy sectors. Passionate about contemporary culture, books, poetry, and stories, she brings depth and insight to her writing. When she’s not chasing a story, she’s likely engrossed in a book or enjoying the company of her dog.
