From Queues to Intelligence: The Evolution of Stream Infrastructure for AI

AI workloads demand more than just scalable infrastructure. They require consistent, high-throughput, and reliable data movement that can support constantly evolving models and use cases. In this talk, we'll walk through how stream infrastructure has evolved inside a modern AI company: from simple durable queues to a sophisticated architecture powering various products and research. We'll cover lessons learned in scaling core systems like Kafka, Flink to support products like ChatGPT and Sora, and how thoughtful design, clear abstractions, and strong observability helped us stay reliable under massive growth. Whether you're focused on infrastructure or AI, this talk will highlight how streaming has become a foundational layer in the AI product stack.