Coherent breaks ground on expanded Sherman, Texas plant backed by NVIDIA partnership and CHIPS Act funding

Coherent said it broke ground on an expanded manufacturing facility in Sherman, Texas, betting that the growth of AI will increasingly depend on domestic optical supply chains, not just faster computer chips. The project is designed to scale output of the indium phosphide (InP) wafers and related laser and optics technologies used to carry data between racks and data centers at high speed.

The company, which makes semiconductor lasers, optical components and compound semiconductors, described the Sherman site as home to what it called the world’s first 6-inch indium phosphide fabrication line. The expansion is intended to increase production of the same type of InP wafers that Coherent says sit at the core of the optical “backbone” of modern AI infrastructure, where indicates must travel farther than copper traces can economically handle.

NVIDIA founder and CEO Jensen Huang attended the groundbreaking and spoke with Coherent CEO Jim Anderson. Huang framed AI as a general-purpose technology that affects industries beyond software, while Anderson positioned the manufacturing milestone as part of reindustrializing the United States. Officials including Sherman Mayor Shawn Temann and Texas Economic Development and Tourism executive director Adriana Cruz also delivered remarks.

The event coincided with new public funding support tied to the U.S. CHIPS Act, a federal effort to strengthen semiconductor manufacturing capacity in the country. Coherent said it is announcing a $50 million CHIPS Act grant to help finance the expanded Sherman facility. The company also pointed to roughly $17 million in earlier support from the Texas CHIPS program and the Sherman Economic Development Corporation.

The push for optical connectivity comes as AI systems scale in size and power. Coherent and NVIDIA used NVIDIA’s Vera Rubin Ultra NVL576 system as an example of how hundreds of GPUs can be linked into a single 576-GPU domain, a scale where copper interconnects become less practical. NVIDIA said in the data center, metal traces that work at shorter distances start to force more announcement-conditioning steps, increasing power draw and adding complexity.

According to the briefing at the groundbreaking, silicon photonics and optics become the more energy-efficient path once systems span multiple racks and longer physical distances. NVIDIA said optics involve a one-time power “penalty” when switching between electrical and light-based indicates, but that distance is then “nearly free,” making optical indicating a better fit for large AI deployments like NVL576. Coherent said its InP-based lasers are designed to support that model of optical interconnects.

Coherent and NVIDIA also highlighted their commercial relationship as an additional driver of capacity expansion. NVIDIA said it is investing $2 billion in Coherent to support research and development, future capacity and U.S.-based manufacturing, and it added that it has a multiyear strategic partnership that includes a multibillion-dollar purchase commitment for advanced laser and optical networking products. Coherent said the Sherman expansion reflects that demand for the “connective tissue” that lets AI systems scale, not just compute.

The manufacturing details matter because the company argues that scaling up output is partly a problem of wafer size and yield. Coherent and NVIDIA said silicon fabs typically use 12-inch wafers, while most global InP production remains on smaller 3- and 4-inch wafers, limiting yield and the number of components per run. Moving to 6-inch wafers, they said, increases usable area substantially, lowering cost and unlocking more volume for the optical networking demand tied to AI.

Not all specifics were laid out publicly. The companies did not provide a timeline for when expanded capacity will begin shipping at full rate, nor did they quantify expected changes in Coherent’s near-term revenue, gross margin, or customer qualification progress. The scope of the multiyear purchase commitment was described broadly as “multibillion-dollar,” without company-by-company breakdowns, and the article did not disclose additional terms tied to the CHIPS grant beyond the amount announced at the event.

Over the next several quarters, investors and customers will likely look for confirmation that the expanded InP line meets yield targets and that Coherent’s output scales in step with NVIDIA’s next-generation networking needs. Watch for updates on production ramp, customer acceptance of optical modules and lasers, and any expanded disclosures tying U.S. capacity additions to shipments of optical components used in NVIDIA switches and data center rack-to-rack connectivity. The broader question is whether optics can scale domestically fast enough to keep AI infrastructure from bottlenecking on connectivity as system sizes continue to grow.