Project Suncatcher aims to revolutionize AI by deploying solar-powered satellite constellations equipped with TPUs (Tensor Processing Units) in low Earth orbit. The concept leverages the Sun’s abundant energy, potentially generating eight times more power in space than on Earth. This setup minimizes reliance on batteries while aiming to scale machine learning (ML) compute capabilities beyond terrestrial limits. The project’s foundation relies on compact satellite formations connected by high-bandwidth, low-latency optical links. Key challenges include establishing high-capacity inter-satellite communication, maintaining stable satellite formations, ensuring radiation tolerance of TPUs, and addressing the economic feasibility of launches. The satellites need to support tens of terabits per second for effective data distribution, achievable through advanced optical technology. Close satellite proximity is essential to enhance signal strength, and initial tests have demonstrated promising results with data transmission rates of 1.6 Tbps. The project also addresses orbital dynamics using advanced physics models to manage tightly clustered satellites within sun-synchronous orbits. Testing of Google’s Trillium TPUs indicates they can withstand space radiation better than expected, with no significant failures up to a tested dose of 15 krad(Si). Historically high launch costs have hindered similar projects, but projections suggest costs could drop to below $200 per kilogram by the mid-2030s, making space-based data centers more economically viable. Next steps include a partnership with Planet to launch prototype satellites in early 2027. This mission will evaluate the operational performance of the satellites and the effectiveness of optical communication links in space. Future developments may involve integrating new computing architectures with a focus on solar power, computing, and thermal management, pushing the boundaries of what's achievable in space computing.