Crossing the Threshold to Dependable Open Source AI Serving
Red Hat, Tuesday, October 20th, 2026
Red Hat CTO for AI Brian Stevens reflects on vLLM and llm-d reaching dependable, production-grade open source AI serving.
Red Hat's Brian Stevens, SVP and CTO for AI, recalls that when open weight models first emerged, the industry lacked an open inference serving system that could run them efficiently.
His team redirected its energy to the vLLM project, which grew with the community into the de facto inference server and is now part of the PyTorch Foundation.
He describes how Red Hat continued that work toward distributed, production-scale serving and argues open source AI serving has now crossed the threshold to dependability.