NVIDIA and Coherent have officially begun construction on a $2 billion optical interconnect manufacturing facility in Texas, with substantial backing from the Department of Commerce's CHIPS and Science Act funding. The facility will produce optical components essential to high-speed data movement within large-scale AI data centers—a supply chain segment that has become a genuine constraint as hyperscalers ramp GPU deployments. Coherent will lead manufacturing operations while NVIDIA provides design specifications and technical integration expertise. The exact allocation between CHIPS Act grants, private capital, and equity stakes has not been fully disclosed, but the project represents a rare instance of NVIDIA investing directly in downstream semiconductor component production rather than relying on third-party suppliers. Texas was selected over competing sites including Arizona and Ohio due to existing infrastructure, workforce availability, and regional utility capacity—a critical consideration given the facility's manufacturing energy demands.
The optical bottleneck is no longer theoretical. In 2024, multiple hyperscalers deploying NVIDIA Blackwell and Hopper clusters experienced genuine supply constraints on 800G and 1.6T coherent optical transceivers needed to interconnect GPU pods. One major cloud provider's AI cluster expansion was delayed by four months due to insufficient optical module inventory, forcing them to stage GPU deployments serially rather than activate data center pods in parallel. Coherent's new Texas plant targets initial production within 24-30 months, with capacity ramping to support 100,000+ GPU systems annually by 2027. This addresses a critical gap: while TSMC, Samsung, and Intel have scaled advanced node capacity for GPU and CPU production, optical component manufacturing remains concentrated in Asia with limited domestic U.S. capacity, creating geopolitical and supply-chain risk for defense and commercial deployments alike.
The move positions NVIDIA to compete more directly with vertically integrated rivals. Intel has been developing its own optical interconnect capabilities through internally-funded R&D, while AMD relies on Xilinx acquisition assets and external partners. This Texas facility gives NVIDIA control over a historically neglected but now mission-critical component layer. Production is expected to begin in late 2026, with ramp-up through 2027-2028. Beyond volume, the facility will serve as a integration testbed, allowing NVIDIA engineering teams to co-design optical modules specifically optimized for Blackwell-generation and next-generation GPU clustering standards. The CHIPS Act component reduces private capital exposure while signaling federal confidence in domestic AI infrastructure resilience—a broader theme as U.S. policy increasingly treats compute supply chains as strategic assets alongside semiconductors.