According to inside sources cited by the Korean publication Chosun Biz, Nvidia CEO Jensen Huang intends to unveil the subsequent generation of AI computing chips, codenamed Feynman, at the GTC 2026 conference. This is reportedly the technology described as “something the world has never seen before.” The Feynman architecture is set to succeed the present Blackwell and Vera Rubin solutions, with its primary advancement being the adoption of TSMC’s cutting-edge A16 (1.6-nanometer) process technology.
One of the most eagerly anticipated features within Feynman is the incorporation of an LPU (Language Processing Unit) hardware stack sourced from Groq. Furthermore, Nvidia is slated to employ a hybrid bonding technique, reminiscent of the 3D cache stacking utilized in AMD’s X3D processors. Placing the LPU modules directly within the GPU package is expected to deliver unparalleled operational speeds for AI agents.
Nvidia is anticipated to provide a comprehensive deep dive into the new architecture’s capabilities and its manufacturing roadmap during GTC 2026. The conference is scheduled to commence on March 15th.
About the Author
