The next battleground in the U.S.-China technology war may not be an AI chip. It may be the tiny optical transceivers that quietly connect thousands of those chips inside the world’s largest data centers.
According to Reuters, the Trump administration and the Federal Communications Commission are preparing restrictions on Chinese-made optical transceivers over concerns they could be used for surveillance, malware or other attacks on critical U.S. infrastructure. No formal rule has been published, but officials are reportedly aiming to introduce new restrictions before the end of the year, expanding a campaign that has already reached Chinese drones, routers and other connected hardware.
On the surface, the proposal is about national security. In reality, it strikes at one of the most concentrated supply chains in the AI economy.
Optical transceivers are the small pluggable modules that move data between GPUs at extraordinary speeds, making them indispensable to modern AI clusters. China dominates that market. Innolight and Eoptolink together account for roughly half of global shipments, supplying many of the high-speed optical links that underpin Nvidia– and AMD-based AI systems. Removing them from the U.S. market would not simply mean finding new vendors. It would mean replacing a manufacturing ecosystem that has taken years to build.
That prospect immediately sent investors searching for alternatives. Coherent, the world’s second-largest transceiver supplier, is widely seen as the biggest beneficiary if Chinese competitors are shut out of future U.S. deployments. Applied Optoelectronics, Fabrinet and, to a lesser extent, Lumentum could also gain market share, while Japanese and Taiwanese optical component makers would likely see increased demand as customers diversify their supply chains.
The catch is that the industry is already operating at full speed. Demand for 800G and next-generation 1.6-terabit optical modules is outpacing supply, and there simply is not enough non-Chinese manufacturing capacity to replace the industry’s two largest suppliers. Hyperscale cloud providers have spent years qualifying secondary vendors in anticipation of geopolitical tensions, but qualifying suppliers is far easier than building factories capable of replacing half the world’s production.
Even Coherent’s apparent advantage comes with an asterisk. Although the company is headquartered in the United States, a substantial portion of its manufacturing footprint remains in China. Whether the proposed restrictions ultimately apply to Chinese-owned companies, products manufactured in China, or both could significantly reshape the competitive landscape.
The disruption would not stop with transceiver manufacturers. Every optical module depends on digital signal processors, lasers and a network of highly specialized component suppliers. Companies such as Marvell, Broadcom, Credo Technology and MaxLinear could see demand soften if Chinese transceiver production declines, while laser manufacturers including Lumentum and Coherent could ultimately benefit as production shifts toward non-Chinese suppliers.
The deeper challenge, however, lies even further upstream. High-performance optical networking depends on indium phosphide, or InP, a compound semiconductor used to manufacture the lasers and silicon photonics devices that make ultra-fast optical communication possible. Here, too, China occupies a critical position. US company AXT Inc, one of the world’s leading suppliers of indium phosphide substrates, produces a significant share of its crystal material in China. The company therefore sits at the center of an uncomfortable paradox. If optical manufacturing migrates out of China, demand for indium phosphide substrates is likely to rise. But if Chinese production or exports become restricted, the same Western manufacturers expected to benefit could find themselves competing for a constrained supply of the material they need most. The issue is particularly relevant for Lumentum, which recently signed a long-term indium phosphide supply agreement with AXT, underscoring just how strategically important these substrates have become.
That highlights the central dilemma behind the proposal. Restricting Chinese transceivers could reduce one dependency while exposing another. Instead of solving the supply chain problem, it may simply move the bottleneck from finished networking hardware to the semiconductor materials that enable it.
In the near term, that could make AI infrastructure more expensive before it makes it more secure. Cutting off the world’s largest transceiver suppliers would tighten an already constrained optical ecosystem, driving higher prices for networking components just as hyperscalers are accelerating investments in ever larger GPU clusters.
There is also an irony to the debate. Executives at companies such as Lumentum and Coherent have spent years arguing that Chinese optical vendors have enjoyed broad access to U.S. cloud customers while American suppliers have faced far greater restrictions in China. From their perspective, the administration’s proposal is less about protectionism than correcting a long-standing competitive imbalance.
Whether Washington ultimately follows through remains uncertain. The White House and the FCC have yet to publicly confirm the proposal, while Beijing has warned it will respond to any measures that materially harm Chinese interests. But the broader trend is unmistakable. The AI race is no longer being fought only over processors. It is spreading into every layer of the hardware stack, from optical modules and lasers to the semiconductor crystals they are built on. In the emerging technology cold war, even the smallest components have become strategic assets.
