Ayar CEO: ‘Copper Is Already Broken; Agentic AI Will Require Optical I/O’

By Sally Ward-Foxton, EETimes (October 7, 2024)

SAN JOSE, California—At the AI Hardware Summit, Ayar Labs CEO Mark Wade presented the results of a sophisticated simulation of large-scale inference systems built by the company to assess what will be required to make next-generation versions of generative AI models economical.

The results show that optical die-to-die interconnects, like Ayar’s technology, will be required to scale LLM inference for next-gen models in a way that makes economic sense. The simulator also provides a useful way to assess the current state of so-called “tokenomics” (the economics of providing and building products on LLM inference APIs).

Click here to read more ...