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

Inference serves as the fundamental engine of AI value, yet the current reliance on frontier models creates significant cost and defensibility challenges for scaling businesses. Companies are increasingly shifting toward custom, post-trained open-source models to reduce inference costs by 70-90% and retain ownership of proprietary user signals. Base10 facilitates this transition by providing a multi-cloud infrastructure layer that optimizes performance and reliability for high-growth applications like Abridge and WhisperFlow. As compute scarcity intensifies, the industry faces a critical need for standardized, modular data centers to industrialize hardware access. Tuhin Srivastava, CEO of Base10, emphasizes that owning the intelligence stack—rather than relying solely on frontier labs—is essential for long-term viability, as the demand for inference continues to grow exponentially across global markets.