Nvidia tipped to be one of the first customers for TSMC's A16 node

Taiwanese news outlet Ctee reports Nvidia might beat Apple to become one of the first companies to use TSMC's next-gen A16 node. Apparently, it aims for a node advantage over AMD's upcoming 2nm chips, a speculation that was first floated by YouTuber Moore's Law is Dead, who stated that some Zen 6 SKUs would have their CPU tiles manufactured on TSMC's N2X process. That isn't entirely implausible because AMD's soon-to-be-announced Epyc Venice CPU has already taped out on an unspecified N2 node.

In recent history, Nvidia has typically stayed one node behind TSMC's cutting-edge technology, with the RTX 50 series opting for a custom 4NP node instead of one from the TSMC N3 family. That was earmarked for Nvidia's upcoming Rubin chips, which have already taped out. Next year's RTX 60 series GPUs will also likely stick with 3N or 3NP. 

The TSMC A16 is the first node to introduce backside power delivery, a feature already available in Intel's 18A. It allows for greater power efficiency thanks to the power lines being moved to the back of the wafer. Additionally, TSMC claims A16 will provide an 8-10% speed improvement along with a 15-20% power reduction at the same speed, and up to 1.10X increase in chip density.

An earlier product roadmap (H/T Tom's Hardware) states it is expected to enter mass production sometime in late 2026. Therefore, one can reasonably assume that any product made on TSMC A16 won't launch until mid-to-late 2027. It will probably be used for Nvidia's 2028-bound Feynman chips shown off earlier this year.