Google LLC today unveiled Project Suncatcher, an ambitious satellite initiative aimed at deploying artificial intelligence chips into orbit.
One of the key motivations behind the project is the significantly higher power output achievable by solar panels in space compared to those on Earth. Additionally, batteries—which are essential on Earth to supply backup power during nighttime or inclement weather when solar generation drops—are unnecessary in space due to continuous sunlight exposure.
Google believes that the superior energy generation capacity of space-based solar panels could eventually offset the high costs associated with launching chips into orbit. Travis Beers, Senior Director of Google’s AI Research division, noted, “When measured on a per-kilowatt-per-year basis, the combined launch and operational expenses of a space-based data center could become comparable to the energy costs of a terrestrial counterpart.”
A major technical hurdle in operating orbital AI clusters lies in the fact that large language models typically require coordination across multiple GPUs. This means GPUs mounted on separate satellites must exchange data efficiently and reliably.
To address this, Google plans to implement Free-Space Optical (FSO) communication between satellites—a technology that encodes network traffic into laser beams. However, the company’s proposed FSO system aims to achieve data transfer rates of tens of terabits per second, which would normally demand thousands of times more power than current FSO implementations can provide.
Google proposes mitigating this power requirement by positioning satellites much closer together. Shorter distances between satellites reduce the power needed for data transmission. While this approach solves one challenge, it introduces another: closely spaced satellites must actively avoid collisions and maintain stable orbits without drifting into one another.
To manage this complexity, Google has developed a set of physical algorithms to model the behavior of densely packed satellite constellations. Their analysis suggests these challenges are manageable. “Our models indicate that with satellites spaced only a few hundred meters apart, modest station-keeping maneuvers may be sufficient to maintain a stable constellation within the required sun-synchronous orbit,” Beers explained.
Google also investigated whether its AI chips could withstand the harsh radiation environment of space. The evaluation focused on Trillium, the latest iteration of the company’s Tensor Processing Unit (TPU) series. Findings showed that even Trillium’s most radiation-sensitive component—its HBM memory—is capable of operating reliably in orbit for multiple years.
