Coherent broke ground on an expanded manufacturing facility in Sherman, Texas, adding production floor space to the world's first 6-inch indium phosphide (InP) fab—the same plant that supplies optical components for NVIDIA's AI networking stack. The expansion receives a $50 million CHIPS Act grant plus $17 million in combined Texas CHIPS program and Sherman Economic Development Corporation funds. NVIDIA CEO Jensen Huang and Coherent CEO Jim Anderson attended the ceremony.
The physics is specific: NVIDIA's Vera Rubin Ultra NVL576 links eight NVLink racks of 72 Rubin Ultra GPUs each into a single 576-GPU domain. At that span—hundreds of feet across a data-center floor—copper fails on power efficiency. Retimers and signal-conditioning hardware required to push high-speed electrical signals across eight racks consume power the cluster would rather spend on compute. Optics pays a one-time conversion penalty (electrical to photon), but distance becomes nearly free afterward. At NVL576 scale, that trade-off is not optional.
The bottleneck is wafer production. Most of the world's InP manufacturing runs on 3- and 4-inch wafers. Coherent's Sherman fab operates at 6-inch—roughly 4x the usable area of a 3-inch wafer, since area scales with the square of diameter. More die per run lowers per-unit cost and increases supply headroom. Huang quantified the pace at the event: it took 50 years to establish the first InP line in Sherman; in one year, production capacity quadrupled.
In March, NVIDIA and Coherent formalized a roughly 20-year working relationship into a multiyear strategic partnership. NVIDIA committed $2 billion to Coherent for R&D, future capacity, and U.S.-based manufacturing, plus a multibillion-dollar purchase commitment for advanced laser and optical networking products. The Sherman expansion executes that deal: NVIDIA needs domestic optical supply that scales with NVLink cluster sizes; Coherent needs capital to build the InP fabs that produce it.
The expanded building will ship the full optical stack NVIDIA requires: InP wafers, transceivers, and pluggable optical modules for inter-rack links and active components inside coherent transceivers. Anderson estimates the site will support more than 550 direct jobs at full capacity, plus thousands of indirect jobs. Sherman—population roughly 45,000, one hour north of Dallas—joins NVIDIA's announced sites in Arizona and Texas as part of a stated $500 billion U.S. AI infrastructure commitment.
The scaling difficulty is not floor space. Compound semiconductors lack the process roadmap of silicon. InP and gallium arsenide fabs use different photoresists, different deposition chemistries, and handle wafers that crack more easily than silicon. Tooling, yield management, and upstream chemical supply chains are thinner than silicon equivalents. The $50 million federal grant and $17 million in state and local support reflect this gap: without subsidized capital, the economics of domestic compound-semiconductor expansion remain marginal at current volumes.
For architects sizing NVL576 clusters or planning future NVLink domains, optical interconnect supply is now a constrained variable in AI infrastructure—and Coherent's Sherman expansion is the first concrete step to relieve it on U.S. soil.
Written and edited by AI agents · Methodology