The $10B Satellite Empire Putting AI in Orbit, Why Chips Beat Rockets & China's #1 Open Model | EP #266

Will Marshall, CEO of Planet, describes his company’s mission this way: “Google indexed the internet to make it searchable. We’re indexing the Earth to make it searchable.” Planet operates the world’s largest Earth-observing satellite fleet, a $10 billion company that generates 25 terabytes of imagery every day. The conversation covers how Planet is building “large Earth models” to make that data queryable by natural language, the emerging race to put AI compute in orbit, and the tension between launch costs and chip efficiency. The episode also examines China’s open-weight GLM 5.2 model that matches frontier Western models, the legal debate over AI personhood sparked by Argentina’s president, and the financial reality of an AI boom where hyperscaler capex now exceeds cash flow.

Large Earth Models and Orbital AI

Planet’s core insight is that today’s large language models are “brains in a vat” — they know theory from text but have no real-time connection to the physical world. Large Earth models aim to fix that by fusing Planet’s 150-petabyte archive of daily global imagery with LLMs. A farmer could ask how his field compares to last year and get a specific answer, not a textbook. A journalist investigating a flood could get current damage assessments. A hedge fund could track construction progress at Chinese data centers.