NVIDIA Expands Indium Phosphide Fab, Betting on Photonics for AI
NVIDIA has broken ground on an expansion of a chip materials facility owned by II-VI, the company it invested in this spring, according to The Associated Press, extending a partnership aimed at securing the components NVIDIA needs for its AI systems. The project expands a wafer fab that began production last year and houses what the companies describe as the world’s first mass-production line for six-inch indium phosphide wafers. When the expansion is complete, the facility’s capacity will quadruple, according to the announcement, giving NVIDIA a much larger supply of the specialty material that optical communications depend on.
The expansion is the latest step in a relationship NVIDIA formalized in March, when it invested $2 billion in II-VI, roughly 13.6 billion yuan, and signed long-term purchase and capacity agreements. The investment was one of NVIDIA’s largest supplier deals, and it signaled the company’s intent to secure components that were becoming bottlenecks for its data center business. This week’s groundbreaking converts that investment into physical capacity, a sign that the partnership is moving from agreement to production.
Indium phosphide is not a household name, but it is becoming a critical material in AI infrastructure. It is used to make the lasers and photodetectors that move data as light, the optical transceivers that connect servers and the components that will power the next generation of data center interconnects. As AI systems grow, the amount of data moving between processors is becoming as important as the computing power itself, and the companies that build the networking layer are racing to secure the materials it requires.
NVIDIA’s interest is strategic. The company has moved beyond selling processors to selling complete AI systems, including the networking that ties thousands of chips together, and it has identified optical components as a constraint on how large those systems can grow. By investing directly in a materials maker, NVIDIA gains influence over capacity decisions and pricing that it would not have as a mere customer. The partnership is part of a pattern of supplier investments the company has made across its supply chain, from memory to power to packaging.
The technology at stake is the shift to co-packaged optics, in which optical components are integrated directly into the same package as the processor, replacing the electrical connections that currently limit performance. That transition requires precisely the kind of compound semiconductor manufacturing II-VI provides, and the expanded fab is intended to make sure NVIDIA has enough of it when the transition accelerates. Executives at both companies described the expansion as preparation for a demand wave they expect to arrive over the next several years.
The scale of the investment reflects NVIDIA’s confidence in that wave. A fourfold increase in capacity is not an incremental expansion; it is a bet that the market for optical components in AI systems will grow far beyond its current size. The companies did not disclose the cost of the expansion or the revenue it is expected to generate, but the $2 billion investment that preceded it gives a sense of the scale NVIDIA is willing to commit. The expansion also carries risk: if the co-packaged optics transition is slower than expected, or if rival approaches win, the capacity will have been built ahead of demand.
The project also has a geopolitical dimension. Indium phosphide production is concentrated in a small number of facilities, and governments have begun treating compound semiconductor supply as strategically important. NVIDIA’s investment in II-VI’s manufacturing keeps a piece of that supply chain within its own network, reducing its exposure to disruptions elsewhere. The companies said the expanded facility will serve customers in multiple regions, but its primary purpose is clear: to give NVIDIA’s AI systems the optical components they will need, on the schedule NVIDIA sets.
For now, the groundbreaking is a statement of intent as much as a construction event. The expansion will take years to complete, and the wafers it produces will arrive in an AI market that is itself changing rapidly. NVIDIA has placed its bet on a future in which data movement is as valuable as computation, and in which the companies that control the materials control the industry’s pace. The fab expansion is that bet taking physical form. The facility itself is part of a small and specialized corner of the semiconductor industry. Indium phosphide wafers are produced in far smaller volumes than the silicon wafers that make most chips, and the manufacturing know-how is concentrated among a handful of suppliers with decades of experience. The fab that began production last year gave II-VI a six-inch line in a market still largely built around smaller formats, and the expansion now being built multiplies its output. For NVIDIA, the advantage of owning the capacity through investment rather than contract is control: the company can steer production toward the components its systems need most, in the quantities its roadmap requires, without competing with other buyers for the same output.
This article was prepared by Rhino Finance’s editorial team based on public reporting.
