Google Research announced Project Suncatcher, a research initiative exploring how a constellation of solar-powered satellites powered by Tensor Processing Unit TPUs could one day enable large-scale artificial intelligence computation in space. This project is part of early-stage research into scalable AI infrastructure that operates beyond the Earth’s surface by leveraging the continuous solar energy available in orbit.
According to Google, satellites operating in sun-synchronous orbits can collect solar power almost continuously, up to eight times more efficiently than ground-based systems. The proposed design envisions a constellation of small satellites linked by free-space optical connections. These high-speed links distribute machine learning workloads across multiple TPUs in orbit, reducing dependence on ground-based data centers and minimizing environmental impact.
The preprint paper “Towards future space-based, highly scalable AI infrastructure system design” details the system architecture and associated technical challenges. These include maintaining high-bandwidth communications between satellites, managing orbital dynamics in dense formations, and ensuring TPU hardware is radiation hardened. Early laboratory experiments demonstrated optical data transmission rates of up to 1.6 Tbit/s using a single transceiver pair.
The researchers also modeled orbital behavior using the Hill-Clohesy-Wiltshire equation to simulate how a cluster of up to 81 satellites could maintain a stable formation at an altitude of around 650km. These simulations suggest that compact satellite groups separated by only a few hundred meters can remain stable with limited station-keeping operations.
Radiation testing of Google’s Trillium TPU v6e showed that the hardware can withstand radiation levels expected during a five-year mission in low Earth orbit. Only minor performance abnormalities were observed even after exposure to levels significantly exceeding the predicted operational dose.
Our analysis further suggests that falling launch costs could make the deployment of space-based computing systems economically viable within the next decade. If current trends continue, launch costs could fall below $200 per kilogram by the mid-2030s, making orbital computing clusters comparable in cost to ground-based data centers in terms of energy consumption.
Google CEO Sundar Pichai commented:
This is only possible thanks to SpaceX’s significant advances in launch technology.
Elon Musk also joined the conversation and said:
The SpaceX team is amazing. Everything has been done without AI so far, including Starship. The possibilities of AI are unimaginable.
As a next step, Google plans to work with Planet to launch two prototype satellites by early 2027. Following Google CEO Sundar Pichai’s post announcing the study, engineers and AI researchers highlighted the project’s potential impact on large-scale computing and sustainable infrastructure. The broader discussion highlighted how such experiments could reshape assumptions about where and how AI systems will be trained in the future.
Project Suncatcher continues Google’s tradition of experimental research in computing infrastructure, following initiatives such as quantum computing and autonomous systems. Still in the early stages of research, the initiative investigates how future AI computation will evolve beyond Earth-based limitations toward scalable and energy-efficient systems in space.
