NVIDIA and optical networking specialist Coherent have broken ground on a $2 billion manufacturing facility in Texas designed to mass-produce optical interconnect technology—a critical infrastructure layer that has emerged as a limiting factor in scaling AI data centers. The project, backed by CHIPS Act funding, directly addresses one of the semiconductor industry's most pressing challenges: moving data between GPU clusters fast enough to avoid becoming a computational bottleneck. As hyperscalers build larger AI training clusters and deploy inference workloads across distributed systems, the electrical copper interconnects and traditional switching architectures that dominate current data centers are hitting fundamental limits in bandwidth density and latency. Optical interconnects can transmit data at vastly higher speeds over longer distances without the signal degradation inherent in electrical transmission, making them essential infrastructure for next-generation AI factories.
The technical advantage is substantial. Electrical interconnects max out around 800 gigabits per second per lane with significant heat dissipation and distance constraints, while optical systems already operate at 1.6 terabits per second and can scale higher with room-temperature operation and minimal power overhead. Coherent, which has spent over a decade developing photonic integration and silicon photonics platforms, brings proven expertise in manufacturing the transceivers, multiplexers, and chiplets that convert electrical signals from GPUs into optical form for transmission across fabric networks. The company's existing capabilities in high-volume optical component production position it uniquely to transition from specialty applications into the data center mass market. Timeline details suggest the facility targets meaningful production capacity within two to three years, with scaling toward millions of optical modules annually to meet hyperscaler demands. Cost per unit has historically been prohibitive—premium optical interconnects can add 15-20% to data center capital expenditure—but the CHIPS Act subsidy and manufacturing scale are expected to compress pricing and improve adoption curves.
The urgency behind this groundbreaking reflects immediate market pressure. Major hyperscalers deploying clusters of thousands of NVIDIA H100 and incoming Blackwell GPUs are already hitting network saturation points where optical interconnects are no longer optional but necessary for cluster efficiency above certain scales. FERC's recent rulings on large-load interconnection for AI factories, announced simultaneously, further underscore regulatory recognition that data center infrastructure—from power delivery to networking—requires coordinated upgrades. The optical interconnect bottleneck has become one of the few hardware barriers that even NVIDIA's GPU dominance cannot overcome alone; the company needs manufacturing partners like Coherent to build the surrounding ecosystem. This $2 billion bet signals that both industry and government view optical fabric as foundational infrastructure on par with GPU manufacturing itself, essential to realizing the full performance potential of trillion-parameter AI systems and the next decade of AI hardware scaling.