A new report claims Nvidia will use its GTC 2026 keynote to look well beyond Vera Rubin, with next-gen Feynman chips tipped to get their first public outing.
Nvidia boss Jensen Huang has already promised the crowd will see technology “never unveiled before” in his keynote, and the Korean outlet Chosun Biz seems convinced Feynman is the main event.
Details remain thin on the ground, but the report suggests Feynman will be the first Nvidia part built on TSMC’s A16 process, described as a 1.6nm-class node with Super Power Rail.
The claim is that Nvidia will be TSMC’s first, and possibly only, A16 customer during the early high-volume manufacturing phase, because others may need painful architectural revamps.
That is not the only spicy angle floating around, as the report’s chatter drags Groq into the picture. The suggestion is that Feynman could be the first Nvidia platform to adopt Groq LPU units, because latency has become the obsession for anyone flogging accelerated compute.
The theory is a hybrid bonding approach that slots LPU hardware into the package, a move likened to AMD’s stacking playbook for its X3D processors. It sounds neat in a slide deck, but pushing LPUs on the package would make design and manufacturing harder, especially once yields and thermals start getting real.
If the show matches the previous Vera Rubin-style reveal, expect capability claims, a broad architecture tour and some carefully massaged timelines while the kit is still in the pipeline.
The same report pushes production to 2028, with customer shipments drifting into 2029 or 2030, depending on how Nvidia allocates supply.
GTC 2026 starts on 15 March and heads back to San Jose, California, where Huang will no doubt grin through another round of “never unveiled before”.
