No Priors Ep. 96 | With Modal CEO and Founder Erik Bernhardsson

Serverless cloud infrastructure for AI and machine learning requires a fundamental shift in how developers interact with compute resources. Erik Bernhardsson, founder of Modal and former Spotify machine learning lead, emphasizes the necessity of flexible, on-demand GPU access to replace rigid, long-term cloud contracts. By building a custom scheduler and container runtime, Modal enables developers to treat cloud compute with the same speed and ease as local development. Transitioning from training-heavy workflows to inference-centric models requires dynamic capacity management to handle bursty, unpredictable demand. Beyond compute, the evolution of AI-native storage solutions and the integration of physics-based simulations represent critical frontiers for infrastructure. Despite rapid advancements in AI coding tools, human software engineers remain essential as these technologies continue to unlock latent demand for increasingly complex software development.