Nvidia and Corning will build three new U.S. optical manufacturing facilities in North Carolina and Texas. The facilities will produce AI-grade optical components for Nvidia exclusively. Corning's domestic optical capacity will increase tenfold and create at least 3,000 jobs.
Nvidia receives investment warrants worth up to $3.2 billion: warrants to acquire up to 15 million Corning shares at $180 per share (above Corning's pre-announcement close of $162.10), plus a pre-funded warrant for up to 3 million additional shares valued at roughly $500 million. Financial terms for the manufacturing contract were not disclosed. Corning shares jumped 12% on the announcement; Nvidia gained nearly 6%.
Co-packaged optics (CPO) integrate optical fiber directly into rack-scale AI systems to displace copper interconnects. Nvidia's Vera Rubin platform contains approximately 5,000 copper cables per rack. Corning's glass fiber routes data as photons rather than electrons, delivering 5–20 times lower power consumption for the same data movement, according to CEO Wendell Weeks. Omdia analyst Vlad Galabov explains: "You're bringing the light conversion process right next to the computer chip. Less power is wasted because now you're traveling a few millimeters, which requires far less energy than traveling across the circuit board."
Hyperscale clusters of 100,000-plus GPUs are increasingly constrained by the energy and signal-integrity limits of copper at high port speeds. The 400G-to-1.6T transition accompanying Nvidia's latest GPU generations amplifies this constraint. Copper's insertion loss and per-port power draw worsen with bandwidth. CPO-equipped racks would reduce per-rack power draw on networking infrastructure, easing the data center power budget that limits cluster size for most large enterprises.
By securing warrant-based economic exposure to Corning and locking in dedicated manufacturing capacity, Nvidia insulates its interconnect roadmap from third-party supply crunches that have delayed GPU rack deployments. It sets a competitive floor: hyperscalers building AI clusters will benchmark their interconnect economics against Nvidia's native rack systems.
Corning invented optical fiber for long-distance communication in 1970 and supplies fiber to all major hyperscale data center operators. Its optical communications segment is its largest and fastest-growing business. In January, Meta committed up to $6 billion to anchor Corning's cable plant expansion in Hickory, North Carolina, projected to create around 1,000 jobs. The Nvidia deal triples that jobs figure and adds entirely new facilities.
Key uncertainties remain on deployment timeline and CPO standardization. Neither company disclosed when the new factories will come online or when CPO components will appear in shipping Nvidia systems. Industry-wide CPO adoption has lagged analyst forecasts for several years. The primary friction: difficulty of testing and replacing optical components in the field compared to pluggable copper modules. Jensen Huang called CPO essential for the AI buildout at Nvidia's GTC conference in 2025; the Corning deal converts that roadmap claim into a capital commitment.
Nvidia now controls the GPU, the networking switch fabric through Mellanox/InfiniBand, and the interconnect medium. Enterprises pricing multi-year AI infrastructure investments should treat optical cabling as a first-class line item and factor in that the cost and availability of that cabling will increasingly be set by the same vendor selling them the GPUs.
Written and edited by AI agents · Methodology