Coherent will quadruple output of indium phosphide wafers, the semiconductor at the core of optical interconnects that AI clusters now need as copper wiring runs out of room.
The supply of indium phosphide, a compound semiconductor that turns electrical signals into laser light for the wiring inside AI data centers, just became a strategic concern for the world's most valuable chip company. Coherent, the optics supplier Nvidia took a $2 billion equity stake in earlier this year, said it will quadruple output of the material at its fab in Sherman, Texas, betting that the next bottleneck in the AI buildout is not compute silicon but the optical plumbing that ties thousands of accelerators together.
The expansion commits $650 million in capital, effectively doubles the plant's footprint, and is expected to create about 1,000 jobs, roughly 550 of them in advanced manufacturing, engineering, and technical roles, according to The Register's reporting on the plan (Coherent to quadruple indium phosphide wafer output in Texas). Coherent did not disclose when the new lines will be online.
The Register's coverage frames the underlying shift as a moment AI clusters move from a few dozen accelerators to racks holding hundreds or thousands of them. At that scale, the copper traces that carry signals between chips run out of bandwidth and reach, and optical links take over, with lasers and photodetectors converting electrical signals into pulses of light that scale further with less heat and power loss. Those links depend on a tight stack of components built around indium phosphide wafers. Coherent operates eight US wafer fabs producing the semiconductors used in laser light sources and optical modules, the building blocks of the optical interconnect stack (The Register, Coherent coverage).
The funding behind the Sherman build shows how seriously Nvidia is treating the bottleneck. Beyond the $2 billion equity investment announced in March 2026, Coherent's expansion draws on $20 million from the Texas Semiconductor Innovation Fund and the Sherman Economic Development Corporation, plus up to $50 million from the federal CHIPS and Science Act, the 2022 US law that subsidizes domestic semiconductor manufacturing (The Register, Coherent coverage).
The Texas project is one piece of a much larger hedging campaign. In roughly 90 days, Nvidia has committed about $2 billion each to three different optical interconnect suppliers: Coherent, Lumentum, and Marvell, a combined $6 billion wager that the AI infrastructure stack will need far more optical capacity than the current supply chain can produce (Nvidia burns $4B to light up US photonics manufacturing). At Computex in early June 2026, Nvidia chief executive Jensen Huang said the underlying silicon photonics technology would make Marvell the next trillion-dollar company, a striking on-record endorsement of the optics thesis from a CEO whose own company dominates AI compute silicon (The tech that could make Marvell the next trillion-dollar company).
The open question is timing. The Register's reporting flags that Coherent has not provided a completion timeline for the Sherman expansion, and that the article frames demand materialization as conditional, hedged with "when (or if) it materializes." That uncertainty matters because every layer of the optics stack, from indium phosphide wafers to laser packaging to the modulators that drive them, is presently capacity-constrained. A 4x increase in wafer output is the kind of supply unlock the industry has been waiting on, but the unlock only pays off if the AI cluster buildout is still ramping two to three years from now.
What to watch next: a completion date from Coherent, a separate demand check from at least one hyperscaler buyer, and any signal that Lumentum or Marvell is also expanding US wafer capacity to mirror the Sherman bet.