For many years, data center networking relied on a familiar formula. Engineers packed more bandwidth into copper cables, tweaked retimers, and optimized signal conditioning to push data across racks. While that approach served the industry well during the cloud computing era, the explosive scale of modern AI clusters is demonstrating that we are entering a new phase. This new era is driven by the physical limitations of copper and the absolute necessity of optics.
At a recent groundbreaking ceremony at its Sherman, Texas campus, Coherent made it clear that its ambitions extend far beyond just incremental capacity upgrades. The company is actively scaling what it calls the world’s first volume-production 6-inch Indium Phosphide (InP) wafer fab. In doing so, it is building out the optical backbone required to keep next-generation AI infrastructure running at full speed.
Breaking the Copper Bottleneck
When we think about massive AI clusters, the focus almost always lands on raw computing power, specifically the GPUs and specialized logic chips. However, as AI systems grow larger, connectivity becomes just as critical as compute.
During a fireside chat at the event, NVIDIA founder and CEO Jensen Huang put the engineering challenge into perspective. When you scale up to a system like the upcoming NVIDIA Vera Rubin Ultra NVL576, which links eight NVLink racks of 72 Rubin Ultra GPUs into a single 576-GPU domain, copper simply cannot carry the signal across those distances without burning massive amounts of power.
To connect hundreds of thousands of processors spread across a massive data center floor, silicon photonics is no longer a luxury. It is the only viable solution. While converting electrical signals to light introduces a one-time overhead, the subsequent distance traveled is virtually free in terms of power efficiency. At NVL576 scale, optics becomes a baseline requirement.
Moving Beyond Legacy Wafers
The semiconductor lasers that power these optical interconnects are not new, but manufacturing them at the scale the AI industry demands has always been a bottleneck. While silicon manufacturing transitioned to large 12-inch wafers long ago, much of the world’s compound semiconductor production, specifically Indium Phosphide, has remained stuck on legacy 3-inch and 4-inch wafers. This results in lower yields and fewer components per run.
This is where Coherent’s Sherman facility represents a genuine manufacturing leap. By moving to a 6-inch wafer process, the company is roughly quadrupling the usable surface area compared to traditional 3-inch wafers. In the hardware manufacturing world, that scaling factor is what drives down production costs and unlocks the sheer volume necessary to support massive global AI deployments.
Inside the fab, the fundamental lithography, etching, and deposition processes look familiar, but the magic lies in growing these exotic compound-semiconductor layers on an InP substrate to precisely modulate and emit light. These chips end up inside Coherent’s pluggable transceivers. These are the USB-sized modules that plug directly into NVIDIA networking switches, supporting advanced architectures like Spectrum-X and Quantum-X Photonics.
Reindustrializing the Ecosystem
Beyond the immediate hardware implications, this expansion signals a deeper shift in where critical infrastructure is built. Backed by a $50 million CHIPS Act direct grant, alongside state and local funding from the Texas CHIPS program, the project represents a serious effort to secure a domestic supply chain for components that have historically migrated overseas.
The partnership is anchored by a massive multiyear strategic agreement signed earlier this year. This includes a $2 billion investment from NVIDIA into Coherent to support R&D and secure future capacity.
For Sherman, a city of 45,000 just north of Dallas, this translates to over 550 direct high-tech jobs and thousands of indirect ecosystem roles. It is a reminder that while the AI era is defined by sophisticated software and models, it is ultimately anchored by heavy industrial manufacturing, advanced physics, and a lot of hardware muscle.
