Recent reports suggest that NVIDIA’s GTC 2026 keynote is set to unveil groundbreaking advancements, particularly the anticipated launch of next-generation Feynman chips, extending beyond the already known Vera Rubin project.
Feynman Chips: Pioneering the Adoption of Groq’s LPU Units
NVIDIA has a clear vision for its future, aiming to redefine computing over the next decade at this year’s GTC. During his keynote, CEO Jensen Huang has promised to present technologies that have “never been unveiled before.”While speculation surrounding a focus on the Feynman project has surfaced, Korean media source Chosun Biz corroborates this view, revealing that GTC will serve as the platform for Feynman’s first public appearance.
While specifics on the Feynman chips remain limited, they will herald the initial deployment of TSMC’s A16 (1.6nm) technology, which represents a substantial leap in semiconductor innovation. This technology features Super Power Rail (SPR) capabilities along with the smallest node technology currently available. Notably, NVIDIA is poised to be TSMC’s primary customer for A16 during the High Volume Manufacturing (HVM) phase, as broader mobile adoption might follow later due to necessary architectural adjustments.
Moreover, we theorize that the Feynman chips might incorporate Groq’s LPU (Latency Processing Unit) hardware stack, as GPU manufacturers increasingly prioritize minimizing latency. In earlier discussions, we highlighted the potential for NVIDIA to utilize a hybrid bonding technique with LPU units, akin to AMD’s X3D processors. Although detailed architectural specifics remain undisclosed, integrating LPUs with Feynman is a logical progression, even though it may complicate design and production processes.
Widely expected is a presentation format for the Feynman showcase reminiscent of NVIDIA’s previous demonstrations with Vera Rubin, where the capabilities, overall architecture, and mass production timelines of the GB200s were detailed. Looking ahead, production of the Feynman chips is projected to commence by 2028, with customer shipments likely to roll out between 2029 and 2030, contingent on NVIDIA’s operational strategy. For those interested, GTC 2026 is scheduled to take place from March 15 in San Jose, California.