Coherent has announced a 200% expansion of indium phosphide (InP) optical interconnect production at its Sherman, Texas manufacturing campus. The project adds 100,000 square feet of dedicated cleanroom manufacturing space to the existing campus, per Coherent’s official Sherman expansion announcement. Full operational capacity for the expanded production lines is targeted within 36 months of the project’s public announcement.
Credit: NVIDIA / Coherent Corp.
The scale-out is designed to meet demand from Coherent’s longstanding supply partnership with NVIDIA. The company has produced optical components for NVIDIA for more than 10 years, per NVIDIA’s official blog post on the collaboration. This expansion builds on Coherent’s existing InP manufacturing operations at the Sherman campus, which has produced optical components for AI infrastructure prior to the announcement.
Coherent Texas InP AI Interconnect Expansion Triples Annual Output to 1.2 Million Units
The expansion lifts annual InP optical component production from a pre-expansion baseline of 400,000 units per year to a post-expansion capacity of 1.2 million units per year, per Coherent’s official Sherman expansion announcement. This 200% capacity increase is scheduled to reach full operational status within 36 months of the project’s public announcement. The tripled output is targeted at meeting demand for high-bandwidth optical interconnects in large-scale AI data center GPU cluster deployments.
Expanded production will supply NVIDIA Blackwell AI networking hardware
Components produced at the expanded Sherman facility will be integrated into two core NVIDIA AI networking products, per NVIDIA’s official blog post on the collaboration. The first is the NVIDIA Quantum-X800 InfiniBand switch, which supports 800 Gbps per port for large-scale AI cluster networking. The second is the NVIDIA MQFS8000 front-panel transceiver, a form-factor pluggable device built for 800 Gbps data center Ethernet deployments.
Both products are fully qualified for use with NVIDIA Blackwell GPU-based data center systems. Compatible hardware includes the HGX B200 server platform and the GB200 NVL72 rack-scale AI platform. This qualification ensures plug-and-play integration for operators deploying large-scale Blackwell GPU clusters.
InP optical interconnects deliver measurable efficiency gains over copper for AI clusters
Per Coherent’s official InP product specifications and NVIDIA’s collaboration blog post, InP-based optical interconnects deliver up to 10 times the bandwidth of traditional copper interconnects. They also consume 50% less power during operation and generate up to 30% lower thermal output than copper alternatives.
For a standard 10,000-GPU NVIDIA Blackwell AI training cluster, copper-based interconnects require hundreds of electrical retimers to maintain signal integrity across the full cluster footprint. InP optical interconnects eliminate this supporting hardware entirely. This reduces total hardware and cooling costs for operators by removing the need to purchase, power, and cool retimer units.
Domestic InP production cuts U.S. AI infrastructure lead time variability
Overseas InP component lead times currently range from 12 to 18 months, per Coherent’s official Sherman expansion announcement. These timelines delay large-scale AI data center deployments by up to 1.5 years for operators sourcing components from non-domestic foundries. This extends time-to-revenue for new AI service launches by over a year for affected projects.
For example, a U.S.-based AI service provider planning a 10,000-GPU Blackwell cluster deployment would face a 12–18 month delay to component delivery if sourcing InP interconnects from overseas foundries, per Coherent’s announcement. Localized production at the Sherman facility eliminates this delay by providing a domestic, qualified source for high-bandwidth optical interconnects. This cuts lead time variability for U.S.-based AI hardware manufacturers compared to overseas sourcing, reducing time-to-revenue for new AI service launches.
The project is supported by funding from the U.S. CHIPS and Science Act, which has allocated $52.7 billion in total funding for U.S. semiconductor manufacturing and research as of 2024, per the official U.S. CHIPS program website. This aligns with federal efforts to onshore critical semiconductor and advanced component manufacturing.
By bringing InP component production to the Sherman campus, Coherent reduces exposure to global supply chain disruptions that have delayed AI data center buildouts in recent years. This supports more predictable deployment timelines for U.S. AI infrastructure projects.
What is the timeline for Coherent’s Sherman, Texas InP expansion to reach full capacity?
Full operational capacity for the expanded 100,000-square-foot cleanroom production lines is targeted within 36 months of the project’s public announcement. At that point, the facility will produce 1.2 million InP optical interconnects per year, per Coherent’s official Sherman expansion announcement.
What NVIDIA AI hardware uses Coherent’s Texas-produced InP interconnects?
Components from the expanded Sherman facility are qualified for integration into the NVIDIA Quantum-X800 InfiniBand switch, which supports 800 Gbps per port, and the NVIDIA MQFS8000 front-panel transceiver, a form-factor pluggable device built for 800 Gbps data center Ethernet deployments. Both products are compatible with NVIDIA Blackwell GPU-based systems including the HGX B200 and GB200 NVL72 rack-scale AI platforms, per NVIDIA’s official collaboration blog post.
How do Coherent’s InP optical interconnects compare to copper for AI data centers?
Per Coherent’s official product specifications and NVIDIA’s collaboration details, InP-based optical interconnects deliver up to 10 times the bandwidth of traditional copper interconnects. They also consume 50% less power during operation and generate up to 30% lower thermal output than copper alternatives.
For a standard 10,000-GPU NVIDIA Blackwell AI training cluster, InP interconnects eliminate the need for hundreds of electrical retimers required for copper setups to maintain signal integrity. This reduces total hardware and cooling costs for operators by removing the need to purchase, power, and cool retimer units across the full cluster footprint.
How does domestic InP production reduce lead times for U.S. AI infrastructure projects?
Overseas InP component lead times range from 12 to 18 months, per Coherent’s expansion announcement, which can delay 10,000-GPU Blackwell cluster deployments by up to 1.5 years for U.S. operators. Domestic production at the Sherman facility cuts this lead time variability, reducing time-to-revenue for new AI service launches. It also limits exposure to global supply chain disruptions that have delayed AI data center buildouts in recent years.
The facility will hit full operational capacity within 36 months of the public announcement, producing 1.2 million InP optical interconnects per year. These parts deliver 10 times the bandwidth of copper, 50% lower power draw, and 30% less thermal output. They are qualified for integration into NVIDIA Blackwell GPU-based data center networking hardware, including Quantum-X800 InfiniBand switches and MQFS8000 front-panel transceivers, with production supported by U.S. CHIPS and Science Act funding.
Bottom line: U.S. AI infrastructure operators planning 10,000-GPU or larger NVIDIA Blackwell GPU cluster deployments should prioritize sourcing InP optical interconnects from Coherent’s expanded Sherman, Texas facility to avoid 12–18 month overseas lead times that can delay time-to-revenue for new AI service launches by up to 1.5 years.
