Moonshot AI's Yang Zhilin (杨植麟) lands rare spot on Nvidia (英伟达) GTC roster — why it matters

The headline act for a Chinese outsider

Nvidia’s (英伟达) annual GTC developer conference in San Jose will feature an unusual guest: Yang Zhilin (杨植麟), founder and CEO of Moonshot AI (月之暗面), reportedly the only representative from an independent large‑model startup on the organiser’s 2026 guest list. The conference—whose keynote by Nvidia CEO Jensen Huang is watched as an industry barometer—runs around March 16–17, and Yang is slated to speak on March 17 at 11 a.m. local time.

What he will present

Yang will deliver a talk billed “Developer tools and techniques — large language models (LLMs),” in which he will outline Moonshot AI’s engineering approach. It has been reported that he will discuss a Muon optimizer that allegedly doubles token‑learning efficiency for the startup’s Kimi K2.5 assistant and a Day‑0 infrastructure co‑design aimed at maximising training throughput. He will also reportedly delve into AI‑native training and linear‑attention architectures to enable longer‑running agents.

Why Western readers should care

Why does this invitation matter? Moonshot AI (月之暗面) is a small, founder‑led player in a space dominated in China by giants such as Baidu and Alibaba; independent startups gaining visibility at Nvidia’s flagship developer event signal both technical credibility and growing international engagement. Against the backdrop of U.S. export controls and broader tech tensions, access to cutting‑edge hardware and engineering partnerships is a strategic pressure point for Chinese AI firms. It has been reported that media have already dubbed Yang “China’s post‑90s big‑model pioneer,” but the proof will be in the technical details he presents on stage — and in whether independent Chinese LLM teams can translate engineering claims into production at scale.