Wedoany.com Report on Mar 18th, Inter-chip interconnection, a core component of the global AI computing infrastructure, is undergoing a historic transition from "electrical" to "optical." On March 16th local time, at the NVIDIA GTC 2026 conference, company founder and CEO Jensen Huang officially unveiled the new-generation Feynman architecture chip. This marks the first introduction of optical communication technology into inter-chip interconnection, an initiative hailed by the industry as a fundamental transformation in AI computing architecture.
According to NVIDIA's on-site presentation, the Feynman chip integrates optical interconnect interfaces at the chip level, significantly boosting data transfer efficiency between chips while dramatically reducing energy consumption. Compared to traditional electrical interconnects, optical communication-based inter-chip interconnection can reduce communication energy consumption in AI data centers by over 70%. This implies that within ultra-large-scale AI clusters, a substantial amount of electricity previously consumed during data transmission will be saved and redirected towards more valuable computing tasks.
The industry significance of this technological breakthrough lies in its fundamental response to the increasingly severe "power wall" and "bandwidth wall" challenges encountered during AI computing power expansion. As model parameters continue to balloon, the demand for communication between chips grows exponentially, making the bottlenecks of traditional electrical interconnects in bandwidth, latency, and power consumption increasingly apparent. The launch of the Feynman architecture signifies NVIDIA's attempt to reshape the interconnect architecture of AI clusters at the physical level, bringing optical communication down from data center-level connections to the space between individual chips.
Analysts believe that with the establishment of overseas technological pathways and the continuous strengthening of domestic industrial policies, leading optical module manufacturers in the A-share market deeply embedded in the global computing supply chain, along with related companies possessing Co-Packaged Optics (CPO) technology reserves, are poised to be the first to enter a period of performance realization amid the new infrastructure wave of "computing-power synergy." The penetration rate of optical communication technology in AI computing infrastructure is rapidly increasing. From data center interconnects to inter-chip connections, the trend of "optics in, copper out" has become irreversible.
The release of the Feynman architecture not only solidifies NVIDIA's technological leadership in the field of AI computing but also opens up new growth avenues for the entire optical communication industry chain. As the demand for AI computing power continues to be unleashed, optical interconnect technology is expected to become one of the core competitive advantages of next-generation AI infrastructure.
