In the world of artificial intelligence, the saying “the sky’s the limit” is almost literal. With the launch of Project Suncatcher, Google Research is literally breaking new ground by exploring ways to power machine learning at scale in space.
According to Google’s announcement, Project Suncatcher is a research moonshot aimed at “scaling machine learning in space.”
The idea is for an interconnected network of solar-powered satellites powered by Google’s Tensor Processing Unit (TPU) chips to harness the full power of the sun to run compute-intensive workloads far above the Earth’s surface.
It may sound like science fiction, but this is the question Google is asking when it comes to AI infrastructure. What if the next frontier in computing wasn’t a data center in Oregon or the Netherlands, but a constellation orbiting the Earth?
why is this cool
The scale and ambition of AI models continues to expand. These include massive computing demands, tight power constraints, and rising infrastructure costs. Google is exploring whether solar-powered orbital platforms can provide a complementary alternative to ground-based computing by moving certain workloads into space.
There are at least three strategic threads here.
-
Energy and Cooling: Satellites are almost constantly exposed to sunlight (depending on their orbit) and can avoid many of the thermal/cooling constraints faced by ground-based data centers.
-
Latency and Coverage: Global constellations have the potential to provide near-ubiquitous access, especially in remote locations and mobile/edge computing scenarios.
-
Business and Strategic Positioning: With growing interest in the space economy, including satellite internet, remote sensing, Earth observation, and communications, the idea of in-orbit computing represents a bold new layer in infrastructure.
Work to date: Prototypes, partnerships, radiological testing
On Google’s research blog, the team published a preprint paper detailing initial radiation testing of the Google TPU in satellite constellation design, control, communications, and space-related environments, explaining that work is already underway.
Importantly, the next step is a learning mission in partnership with Planet Labs (Planet), with the aim of launching two prototype satellites and testing the hardware in orbit by early 2027.
That timeline offers a glimpse into the seriousness behind this “moonshot.” It’s not just a thought experiment, it’s a practical program with milestones.
Lesson: Bold vision, long view
Project Suncatcher is one of the signature initiatives that combines ambition and infrastructure. It does not promise significant short-term change, but rather suggests a long-term strategy. For entrepreneurs, investors, policy makers, and technologists, the takeaways are clear. The next stage in computing is likely to be a higher trajectory than just large data centers.
As Google says, they are “working backwards from this potential future” (their words).
The question is, who else will follow? Who will build tomorrow’s orbital computational grid?
And when the sun stops setting on AI, will our expectations of computing and connectivity change again?
Fast Company (SA)
