Coherent has started expanding its wafer fabrication plant in Sherman, Texas, as demand rises for optical components used in next generation AI data centers. The facility will increase production of 6 inch indium phosphide wafers, which are used in lasers and optical systems that move data between chips, servers, racks, and data centers.
The expansion matters because AI infrastructure is beginning to outgrow traditional copper based data links. NVIDIA’s future Vera Rubin Ultra NVL576 platform is expected to connect 576 GPUs across eight racks as one large system. Moving data across that kind of platform requires massive bandwidth, low latency, and better energy efficiency than copper links can easily provide at scale.
Coherent is one of the suppliers helping NVIDIA move toward optical fabrics and co packaged optics. These technologies use light to carry data, reducing some of the signal loss, power draw, and cabling limits that come with copper.
Why NVIDIA needs optical links for larger AI systems
Modern AI systems are no longer built around a few GPUs inside one server. They are becoming rack scale and multi rack systems where hundreds of accelerators must work together as if they are one machine.
That creates a major networking problem. GPUs need to exchange huge amounts of data during training and inference. If the links between them are too slow or power hungry, the system loses efficiency.
Copper works well over short distances, but it becomes harder to scale as bandwidth rises and systems spread across more racks. Signal quality drops, cables get bulky, and power use increases. Optical links can carry data farther and faster with less loss once the initial conversion cost is handled.
| Area | Why optics matter |
|---|---|
| AI racks | Connects large GPU clusters more efficiently |
| Power use | Helps reduce losses compared with copper at scale |
| Bandwidth | Supports high speed data movement across systems |
| Cabling | Can reduce bulk in dense data centers |
| NVIDIA platforms | Supports future Vera Rubin class systems |
| Coherent role | Supplies lasers, optical components, and compound semiconductors |
Coherent’s Texas expansion supports indium phosphide wafer production
The Sherman, Texas expansion is focused on scaling indium phosphide wafer production. Indium phosphide is important for optical communication because it is used in lasers and photonic components that help convert electrical signals into light and back again.
Coherent’s products already support NVIDIA networking switches through pluggable optics. These optical modules help move data through systems such as Spectrum X and Quantum X based networking platforms.
The new facility is being supported by a $50 million CHIPS Act grant. It also builds on $17 million in support from the Texas CHIPS program and the Sherman Economic Development Corporation. The project fits into a larger push to expand AI infrastructure manufacturing in the United States.
Co packaged optics could become essential for AI data centers
Co packaged optics move optical components closer to the chips and switches that process data. This can reduce the distance that electrical signals need to travel, improving efficiency and helping high bandwidth systems scale.
For NVIDIA, this is important because the company is building increasingly dense AI systems. A platform such as Vera Rubin Ultra NVL576 needs fast communication across hundreds of GPUs. If networking becomes the bottleneck, adding more GPUs becomes less useful.
Optics help address that issue by giving data a more efficient path across the system. The long term goal is not only faster networking, but better power efficiency and more usable compute per data center rack.
Copper is not disappearing, but its limits are clearer
Copper will still remain part of data center infrastructure. It is mature, widely used, and cost effective in many short reach situations. But AI systems are pushing power and bandwidth requirements to levels where copper alone may not be enough.
That is why companies are investing in silicon photonics, optical fabrics, and co packaged optics. These technologies are becoming part of the AI hardware stack, alongside GPUs, CPUs, HBM memory, advanced packaging, liquid cooling, and high voltage power delivery.
As AI racks grow, the interconnect becomes just as important as the accelerator itself. A powerful GPU cluster needs a network that can keep every chip fed with data.
The AI hardware race is moving beyond GPUs
Coherent’s Texas expansion shows how broad the AI supply chain has become. The future of AI data centers is not only about who makes the fastest GPU. It also depends on who can supply the lasers, optics, wafers, switches, memory, power systems, and cooling hardware needed to run those GPUs at scale.
NVIDIA’s partnership with companies such as Coherent points toward a data center future where light carries more of the traffic inside AI systems. That shift could reduce power pressure, improve cluster performance, and help hyperscalers build larger AI factories.
The new Sherman facility is a manufacturing step toward that future. As NVIDIA prepares Vera Rubin and later platforms, optical networking could become one of the key technologies that makes 576 GPU and larger systems practical.
Discussion (0)
Be the first to comment.