Standardizing the AI stack with PyTorch ft. Steven Pousty

You might view PyTorch simply as a tool for data scientists building models, but it has become a critical infrastructure bridge for modern AI-powered workloads. Acting as the functional "kernel" of the modern AI stack, PyTorch handles the complex mathematical abstractions and hardware offloading necessary to turn raw computing power into scalable enterprise systems.
In this episode of Technically Speaking with Chris Wright, Red Hat Chief Technology Officer Chris Wright sits down with Steve Pousty, Principal AI Community Architect in the Red Hat Open Source AI Program Office. Together, they map out how open source is driving the evolution of data architecture and why Red Hat is focusing heavily on this ecosystem. They draw clear historical parallels to the early days of Linux and Kubernetes, outlining how the enterprise landscape is rapidly transitioning from raw upstream projects to fully supported, standardized AI distributions.The conversation explores the key architectural pillars within the PyTorch Foundation ecosystem.
Chris and Steve break down the importance of distributed computing using cluster systems like Ray, hardware-agnostic model serving powered by vLLM, and optimization frameworks like Helion DSL that slash execution runtime constraints at scale. They also detail the introduction of Kubernetes-native distributed inference engines like llm-d and explain how open source governance establishes a neutral playing field that provides true hardware choice for the data center. Tune in to discover how to transition your AI strategies from lab experiments to robust enterprise deployments.