.
Image: Lionel Bonaventure / AFP.
In the world of artificial intelligence, the phrase “the sky’s the limit” is almost literal. With the launch of Project Suncatcher, Google Research is exposing new horizons—quite literally—by exploring how to power large-scale machine learning in space.
According to Google’s announcement, Project Suncatcher is a research moonshot aimed at “scaling machine learning in space.”
The idea: an interconnected network of solar-powered satellites equipped with Google’s Tensor Processing Unit (TPU) chips, harnessing the full power of the Sun to run compute-heavy workloads far above Earth’s surface.
It might sound like science fiction, but in the realm of AI infrastructure, Google is asking: what if the next frontier of compute isn’t a data centre in Oregon or the Netherlands, but a constellation circling our planet?
AI models continue to balloon in size and ambition: enormous compute demands, intense power constraints, and rising infrastructure cost. By moving certain workloads into space, Google is exploring whether solar-baked orbital platforms might offer a complementary alternative to terrestrial compute.
There are at least three strategic threads here:
Energy & cooling: Satellites get nearly constant exposure to sunlight (depending on orbit) and can avoid many of the thermal/cooling constraints faced by earthbound data centres.
Latency & coverage: A global constellation potentially offers near-ubiquitous access, especially for remote regions or mobile/edge compute scenarios.
Business & strategic positioning: With interest in the space economy booming—satellite internet, remote sensing, Earth-observation, communications—the compute-in-orbit idea represents a bold new layer in infrastructure.
Google’s research blog explains that work is already underway as the team has published a pre-print paper detailing satellite constellation design, control, communication, and initial radiation testing of Google TPUs in space-relevant environments.
Importantly, the next step is a learning mission, in partnership with Planet Labs (Planet), which aims to launch two prototype satellites by early 2027 to test hardware in orbit.
That timeline gives a glimpse of the seriousness behind this “moonshot"; it’s not a mere thought experiment but a working programme with milestones.
Project Suncatcher stands as one of those hallmark initiatives where ambition meets infrastructure. It doesn’t promise near-term sweeping change—it signals a long-term strategy. For entrepreneurs, investors, policy-makers and technologists, the takeaway is clear: the next stage of computing may not just be bigger data centres, but higher orbits.
As Google puts it, they’re “working backwards from this potential future” (their words).
The question is: who else will follow? Who will build the orbiting compute grids of tomorrow?
And when the Sun never sets on AI, will our expectations of compute and connectivity shift again?
FAST COMPANY (SA)