NVIDIA Fast-Forwarded Co-Packaged Optics Five Years Ahead of Schedule, Arriving First With Its Feynman GPUs
TL;DR AI
2 min readKey summary
NVIDIA is reportedly pulling co-packaged optics forward by about five years, with first deployment now targeted for Feynman GPUs in 2028 instead of 2033.
Feynman is said to pair the new optics with 3D die stacking, custom HBM, a new Rosa CPU, and other next-gen data center chips.
The move could boost bandwidth and cut latency in AI data centers as networking and power constraints tighten.
AMD is also reportedly working on similar optics technology, underscoring broader industry interest in silicon photonics.

