1.6nm and LPU on a Single Substrate: Is This a Breakthrough or an Act of Desperation? NVIDIA Might Drop a Bombshell at GTC 2026

The annual NVIDIA GTC 2026 conference kicks off in March, and if leaks from Asian industry sources are anything to go by, we're in for more than just a routine lineup refresh. Jensen Huang is gearing up to turn the semiconductor industry on its head by unveiling the Feynman architecture.

The NVIDIA chief has already hinted that he will showcase tech "the world has never seen." Now, the mystery is slowly unfolding: according to information that has leaked to the press, NVIDIA intends to be the pioneer in adopting the 1.6-nanometer process node.

TSMC's A16 (1.6nm) process isn't just another shrink in technical specifications. TSMC is implementing Super Power Rail technology, which moves power lines to the back side of the wafer, significantly boosting logic cell density and improving energy efficiency. According to sources, NVIDIA will be TSMC's first (and initially, possibly only) customer for this process node. Mobile chip makers, it's claimed, are hitting pause — they'll need to seriously overhaul their solution architectures to adapt to the new routing standards.

But the most intriguing aspect of the Feynman story is its potential symbiosis with the developments of a company called Groq. This American startup has long been championing the idea of LPUs (Language Processing Units) — specialized accelerators focused not on peak performance in teraflops, but on minimal latency when working with AI. Rumors are swirling that NVIDIA might integrate Groq's blocks directly into its massive Feynman chips.

If this is true, we'll be looking at a substrate housing two types of cores at once: classic GPUs for "heavy matrix math" and LPU accelerators for lightning-fast AI responses. According to analyst projections, mass production of the new product won't start until 2028 at the earliest, with the first customers getting their hands on solutions based on the new architecture only by 2029.

What do you think: is the industry ready for a 1.6nm process, or are manufacturers hitting a physical ceiling, with nanometer labeling having long ago become just marketing? Share your opinion in the comments!