Google’s latest “moonshot,” Project Suncatcher, envisions a radical fix for a problem straining global energy grids: constellations of solar-powered satellites with its custom AI chips to explore running machine learning workloads in the vacuum of space.
The project, unveiled in a blog post and detailed preprint paper on November 4, is positioned as a bold foray into overcoming Earth’s energy and scalability constraints.
“The sun is the ultimate energy source in our solar system, emitting more power than 100 trillion times humanity’s total electricity production,” writes Travis Beals, senior director of Google’s Paradigms of Intelligence, who is reportedly leading the project. He notes that in the right orbit, solar panels can be “up to 8 times more productive than on Earth,” providing near-continuous power.
“In the future, space may be the best place to scale AI compute.”
The proposed system involves clusters of satellites in sun-synchronous low-Earth orbit, separated by roughly 100 to 200 metres within a roughly 1 km radius formation. These would be linked by ultra-high-bandwidth optical connections to function as a single, distributed data center. According to the preprint paper, Google has already demonstrated a bench-scale link achieving 1.6 tbps total bandwidth, “validating the potential of this approach,” the authors note.
Daunting Checklist
While the vision is grand, the technical and economic challenges are astronomical.
The paper outlines a daunting checklist.
First, Google tested its Trillium TPU chips in a proton beam and found them surprisingly resilient, with memory subsystems beginning to falter at nearly three times the expected shielded radiation dose for a five-year mission. However, long-term reliability in the tough space environment remains unproven. “While [the] error rate is likely acceptable for inference, the impact of Single Even Effects on training jobs, and the efficacy of system-level mitigations, requires further study,” the paper notes.
Second, the project’s feasibility hinges on a precipitous drop in launch costs. Google’s analysis projects prices falling to “less than $200/kg by the mid-2030s,” a figure that is aspirational and dependent on the continued success and cost-cutting of companies like SpaceX.
Third, maintaining a stable, kilometer-wide cluster of satellites requires precise station-keeping to counteract gravitational perturbations, a complex feat of aerospace engineering on a scale not yet attempted.
Visionary or Quixotic?
Google is reportedly already looking to move from theory to early testing.
The company announced a partnership with Earth-imaging firm Planet to launch two prototype satellites by early 2027 — a “learning mission” that will test the core technologies and the hardware in orbit.
Project Suncatcher seems to be a long-term reach, and perhaps not a solution to the immediate energy constraints of AI data centers. A single hyperscale data center on Earth can consume over 100 MW, and replicating that capacity in orbit would require a constellation of immense scale and cost.
The project shows the industry’s rising hunger for sustainable compute power. As AI models grow larger, the sheer energy required to train and run them is likely to hinder innovation. The quest to scale AI has become so energy-intensive that the tech industry is now forced to look beyond the Earth itself for answers. With this move, is Google essentially arguing that if we cannot power the future of AI on Earth, they may have to plug into the sun itself?
