The architecture marks the first time SpaceX has discussed formation flying and involves a hybrid model of larger 250kW satellites and these tight clusters. To maintain coherence, these clusters must operate in sub 750m proximity, a shift from the kilometer-scale spacing historically used for Starlink. Analysis of the setup suggests it could support coherence for up to 720 GPUs, more than double the approximately 300 GPUs in previously tested DeepSeek clusters.
Tuesday, October 20, 2026
Story
SpaceX Plans 1 Million Starmind Satellites for First AI Compute Formations
AI14 days ago
SpaceX VP Michael Nicolls revealed that the company is developing a method to group satellites into tight clusters to enable the coherent compute required for efficient AI processing. This approach involves grouping approximately 10 satellites with 10 terabits of bandwidth between them, all integrated into a larger planned constellation of 1 million Starmind satellites. The goal is to ensure sufficient interconnected bandwidth and low latency to run large AI models in orbit.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.