Jensen Huang flags photonics bottleneck

Nvidia Chief Executive Jensen Huang has spent the past three months putting a specific constraint on the AI buildout into public view: the network that links chips together. At GTC in March, Huang said the amount of silicon photonics capacity Nvidia needs is “substantially higher than the world has today,” according to CNBC’s account of the event. Since then, Nvidia has announced multibillion-dollar optics partnerships with Lumentum, Marvell, Coherent and Corning as it pushes co-packaged optics and silicon photonics deeper into its AI networking stack. That has given investors a cleaner way to describe the next choke point in AI infrastructure. The first phase of the boom centered on GPUs and power; the current argument from Nvidia and its partners is that optical connectivity, fiber and photonics now sit closer to the center of the scaling problem for larger AI clusters. Nvidia says its silicon photonics switches are designed to connect “millions of GPUs across sites” while cutting energy use and raising resiliency. (cnbc.com) ### Why is Nvidia talking about photonics now? March 18 brought Nvidia’s launch of Spectrum-X Photonics and Quantum-X Photonics, two switch platforms built around co-packaged optics. Nvidia said the products are meant for “AI factories” at million-GPU scale and deliver 3.5x energy savings and 10x resilience versus traditional approaches. Huang said at the launch that “networking infrastructure must be reinvented to keep pace.” (nvidianews.nvidia.com) Photonics replaces part of the copper-and-electricity path with light-based data movement. CNBC reported that analysts see that as a way to lower power use and avoid a “scalability and performance wall” as model size and token traffic rise. Morningstar analyst Brian Colello told CNBC that Nvidia’s next rack-scale systems will require increasing amounts of optical connectivity as bandwidth demand rises. (nvidianews.nvidia.com) ### Which companies are tied most directly to that push? Lumentum said on March 2 that Nvidia agreed to invest $2 billion, alongside a multibillion-dollar purchase commitment and capacity rights for advanced laser components. Nvidia said the partnership would support a new U.S. fabrication facility and expand manufacturing capacity for future AI data centers. (cnbc.com) Marvell said on March 31 that Nvidia invested $2 billion as part of a broader tie-up around NVLink Fusion, custom AI infrastructure and silicon photonics. Marvell CEO Matt Murphy said the partnership reflected the “growing importance of high-speed connectivity, optical interconnect and accelerated infrastructure in scaling AI.” (nvidianews.nvidia.com) Corning said on May 6 that it would increase U.S. optical connectivity manufacturing capacity 10-fold and expand fiber production capacity by more than 50% under a multiyear partnership with Nvidia. Nvidia said the expansion would include three new manufacturing facilities in North Carolina and Texas. (nvidianews.nvidia.com) ### Where does the stock-market reaction come from? CNBC reported on May 29 that Nvidia had committed at least $6.5 billion to photonics-related companies since March. The companies named in that tally included Marvell, Lumentum, Coherent, Corning and Ayar Labs. CNBC also reported that Marvell shares were up 122% in 2026 and Corning shares 111% at that point. Social posts this week linked that spending wave to gains in optics names including Marvell, Corning and Lumentum. (nvidianews.nvidia.com) The more durable fact behind those posts is that Nvidia has moved from talking about optics as an engineering option to signing supply, capacity and manufacturing agreements around it. ### How does HPE fit into the same story? HPE said on June 1 that second-quarter revenue rose 40% from a year earlier to $10.7 billion, with record revenue, gross margin and free cash flow. (cnbc.com) Antonio Neri, HPE’s chief executive, said customers were continuing to invest in modernizing infrastructure and scaling AI. HPE’s networking revenue rose 148.2% year over year to $2.7 billion, and data center networking revenue rose 233.3% to $320 million, the company said. (cnbc.com) Those figures do not make HPE an optics pure play, but they do show that the broader AI infrastructure buildout is still feeding through to servers and networking alongside the photonics push. (hpe.com) ### What should readers watch next? Nvidia’s next concrete milestones are already on the record: its Spectrum-X Photonics and Quantum-X Photonics platforms are listed as production networking products, and the company has named CoreWeave, Lambda, Meta, Microsoft and Oracle Cloud Infrastructure as first adopters of Spectrum-X Photonics for Vera Rubin systems. Corning’s capacity buildout and HPE’s updated fiscal 2026 guidance are the next public markers to watch. (hpe.com) (nvidia.com)