NVIDIA is developing next-generation 'Feynman' GPUs that will use TSMC's advanced 1.6 nm A16 node, bypassing the N2 family entirely. Mass production is slated for the second half of 2028. The Feynman architecture will leverage backside power delivery and innovative packaging technologies including SoIC 3D stacking and CoWoS-L 2.5D integration to achieve multi-kilowatt designs capable of processing dozens of PetaFLOPS.

For memory, NVIDIA plans to develop customized HBM4E with logic elements for data preprocessing. System bandwidth will reach over 1,000 TB/s via co-packaged optics (CPO), surpassing Rubin Ultra's 520 TB/s and entering the petabyte range. All this will utilize TSMC's COUPE technology for optical interconnects.