a16z Podcast: Why Open Source Became AI’s Inference Backbone

Open-source AI isn’t just surviving the frontier race—it’s becoming the infrastructure beneath it. In this condensed summary of the full a16z Podcast episode, Matt Bournestein talks with Simon Mo, co-founder of Infract and lead maintainer of vLLM, about how inference, open-weight models, and AI infrastructure are reshaping the stack. Moving from the original full-length conversation to a minutes-long listen, you’ll learn why open source now matters for deployment, guardrails, cost control, and reliability at scale, and how model labs, Nvidia, AMD, Google, Amazon, and Intel are all co-designing the future of serving AI. The discussion also covers the narrowing capability gap with frontier models, the economics of token spend and SLAs, and why open weights may be the backbone of the next phase of AI innovation. Listen now to get the key ideas in minutes.