San Francisco's scene story today is not a stealth salon or a moody warehouse launch party. It is SGLang Summit 2026, a two day Fort Mason takeover that says the center of gravity has moved from demoing model vibes to comparing inference stacks in public. The official event page promises 2,000 plus engineers, researchers, founders, and builders, with speakers including OpenAI's Lilian Weng, Thinking Machines Lab's Soumith Chintala, Intel CEO Lip-Bu Tan, and Perplexity CEO Aravind Srinivas. That is not a meetup lineup. That is a census of who currently gets waved through the fast lane in applied AI.

The anthropology is in the side quests. The summit's community events calendar frames the week as dinners, meetups, networking, and, bless them, "just plain fun." Meanwhile the workshop menu reads like a dating app for infrastructure people: a Local AI Coding Lab, Nebius on inference optimization, Crusoe on managed Kubernetes, and NVIDIA on hands on RL with SGLang and Miles. A conference where the side rooms are about kernels, rollouts, and seat scarcity is a conference telling you exactly what year it is.

If you squint, the whole thing doubles as a social graph for the post chatbot economy. The main stage is frontier names and power signals. The workshops are where people audition their stack, their cloud credits, and their tolerance for first come first served humility. The city has seen plenty of AI events; this one feels especially revealing because it is less about whether agents are real and more about whose plumbing gets adopted while everyone pretends they came for the talks.

So yes, the scene is alive. It just smells less like launch party prosecco and more like complimentary matcha near a GPU expo.