Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, Applied AI

AI inference serves as the primary engine for delivering value in the artificial intelligence ecosystem, yet the current market remains heavily skewed toward expensive frontier models. Scaling companies increasingly adopt custom, post-trained open-source models to improve gross margins by 70-90% and maintain defensibility over proprietary user signals. Base10 facilitates this transition by providing a multi-cloud infrastructure layer that abstracts compute complexity and optimizes performance for high-volume applications like Abridge and WhisperFlow. Persistent compute scarcity, characterized by 15-month lead times for hardware, forces a strategic shift toward owning physical infrastructure and modular data centers. Establishing a robust open-source ecosystem is critical to preventing the concentration of intelligence within a few dominant labs, ensuring that the cost of intelligence remains competitive and accessible for the broader application layer.