AI Networking and Optics

What’s happening? 

Jensen Huang, the founder and CEO of Nvidia, called Marvell Technology the next trillion-dollar company at Computex 2026. Speaking onstage with Marvell CEO Matt Murphy in Taipei, Huang argued that as AI training and inference scale to massive, disaggregated data centers, the bottleneck shifts from raw compute to how efficiently chips communicate. He said Marvell’s networking and connectivity chips are “so essential” because they enable thousands of accelerators, memory units, and compute nodes to operate as a single, distributed supercomputer.

Seizing the Opportunity

Noonum’s Agent transformed a simple investment objective on this topic of inter-chip communication to its broadest level to include optical connections.  Here is the strategy: Invest in companies that supply the networking and optical connections that link AI chips inside and between data centers, including high-speed Ethernet and InfiniBand switches, network interface cards and data processing units, optical transceivers, co-packaged optics, lasers and photonic components, and high-speed copper and fiber cables. The focus is on firms whose growth is driven by the bandwidth needs of large AI clusters.

Noonum’s Strategy Builder delivered more than 200 companies in this basket, with Coherent Corp, Marvell Technology, and Lumentum Holdings in the top 10. The Noonum basket’s simple return for the TTM is sizzling with a few of its companies returning more than 300%. Of course, there is a lot of volatility, as expected from such an emerging opportunity.

Key Insights

Bandwidth is the real organizing principle. The constituent set is not simply a list of 'AI stocks.' It concentrates on the hardware that moves data between chips, servers, racks, and data centers: switches, NICs, DPUs, transceivers, photonic engines, lasers, connectors, and dense cabling. That makes the index more specific than a broad semiconductor basket.

The theme is global, not U.S.-only. Company presence is heaviest in Taiwan and China, with the United States, Hong Kong, Japan, and South Korea also strongly represented. Much of the AI interconnect value chain sits in Asian optical, connector, and manufacturing ecosystems, while U.S. names contribute major platform and systems leadership.

Optics is the center of gravity. The most repeated inclusion pattern is not generic networking software or cloud services. It is optical: 800G and 1.6T transceivers, silicon photonics, indium phosphide devices, CW and EML lasers, modulators, photodiodes, and optical engines. That points to optical scaling as a central bottleneck and opportunity in AI cluster buildouts.

More information at https://noonum.ai/blog/ai-networking-optics‍