Chinese Optical Transceivers Sit at the Heart of America’s AI Data Centers as Congress Targets a New Supply-Chain Risk
A new bipartisan bill in the U.S. Senate is drawing attention to one of the least visible but most important pieces of America’s artificial-intelligence infrastructure: optical transceivers made by Chinese companies. These small devices convert electrical and optical signals so enormous volumes of data can move rapidly through fiber-optic networks inside AI data centers. Modern AI clusters can require hundreds of thousands or even millions of optical modules, making them essential to the networks connecting GPUs, servers, storage systems and cloud infrastructure.
The legislation, introduced by Senators Dave McCormick, Ruben Gallego, John Cornyn and John Fetterman, would prohibit certain Chinese-made optical transceivers from being procured for sensitive federal national-security systems. The measure specifically identifies products from China’s Zhongji Innolight and Eoptolink, along with subsidiaries and affiliates, while allowing the Commerce and Defense departments to designate additional covered suppliers. It would provide a five-year transition period and limited waivers when trusted alternatives are unavailable.
The security concern is rooted in where these components sit inside modern networks. Optical transceivers are not peripheral accessories. They are positioned directly along the high-speed pathways carrying information between machines. Advanced models can include reprogrammable firmware and embedded electronics, creating a supply-chain issue different from buying an ordinary passive cable. Supporters of the legislation argue that foreign-controlled components inside sensitive communications and AI infrastructure deserve scrutiny because interference, compromised firmware, defective updates or supply disruption could affect access to critical systems.
That concern becomes more significant because Chinese companies have become major suppliers of optical transceivers used in data-center infrastructure. Reuters reported that Chinese vendors dominate important parts of the market, while U.S. hyperscale cloud operators are rapidly expanding AI facilities that consume enormous numbers of these modules. Innolight and Eoptolink have emerged as particularly important vendors as bandwidth requirements jump from 400G and 800G toward even faster networking architectures.
Innolight also carries an additional national-security designation. The Pentagon added the company to its Section 1260H list of companies it identifies as connected to China’s military ecosystem. The designation itself does not automatically ban every commercial transaction, but it places the company inside a broader U.S. framework examining links between Chinese civilian technology suppliers and Beijing’s military-industrial system.
For the United States, the larger issue is dependency. AI models attract most public attention because they generate text, images, software and other outputs, but frontier AI depends on vast physical infrastructure underneath them. Thousands of accelerators must constantly exchange data. That traffic moves through switches, fiber and optical transceivers. If one country dominates a component that every advanced data center requires in enormous quantities, that supplier base can become a strategic chokepoint.
The potential vulnerability is similar to problems already seen in semiconductors, rare earth minerals, batteries and telecommunications equipment. A component may appear commercially routine until demand surges or geopolitical restrictions suddenly make replacements difficult. AI infrastructure magnifies this problem because large clusters are built at extraordinary scale. A hyperscale project may need quantities that smaller domestic suppliers cannot immediately replace.
That supply constraint is also the strongest argument raised against moving too quickly. U.S. industry groups have warned that American alternatives such as Coherent and Lumentum currently lack the manufacturing scale needed to replace Chinese vendors across the entire AI buildout. The Information Technology Industry Council urged regulators in August to consider how supply disruptions could slow data-center deployment and weaken U.S. AI competitiveness.
This creates a difficult strategic tradeoff. Reducing Chinese components in highly sensitive government networks can lower exposure to foreign supply-chain risk, but removing dominant suppliers too rapidly could also delay the expansion of the very AI infrastructure the United States is trying to strengthen. That tension helps explain the proposed five-year transition period rather than an immediate prohibition. The bill also directs the Commerce Department to assess U.S. and allied production capacity and develop a strategy for expanding trusted supply.
The debate therefore centers on sequencing as much as security. If replacement capacity is too small, restrictions can create bottlenecks. If diversification is postponed indefinitely, dependence becomes harder to reverse. The practical challenge is building enough American and allied manufacturing capacity before Chinese-origin transceivers become even more deeply embedded in critical AI and cloud systems.
The scale of the market makes that challenge significant. Optical networking demand is rising rapidly because each new generation of AI hardware requires more bandwidth between machines. The performance of an AI cluster is limited not only by how fast individual GPUs can calculate, but by how quickly thousands of processors can exchange data. Optical transceivers therefore become increasingly important as AI systems scale across entire data-center campuses.
That means supply-chain security must move beyond the chips themselves. The United States has spent years scrutinizing advanced processors, semiconductor manufacturing equipment and memory technology, yet an AI system also depends on networking components, power systems, cooling infrastructure, firmware and software. A secure accelerator surrounded by compromised or unavailable infrastructure still creates operational risk.
The federal-government focus is narrower than a nationwide commercial ban. S. 5548 concerns procurement for national-security systems rather than immediately removing every Chinese optical transceiver from private American data centers. That distinction matters. The legislation targets the environments carrying some of the government’s most sensitive communications while allowing the broader commercial market more time to diversify.
The debate nevertheless sends a signal well beyond federal procurement. Major cloud providers, AI developers and networking companies now have reason to evaluate where optical components are manufactured, who controls the firmware, how updates are delivered, what second-source suppliers exist and whether replacement capacity can be obtained quickly during a geopolitical disruption.
The same supply-chain logic that reshaped telecommunications procurement is now reaching AI infrastructure. Huawei and ZTE made policymakers familiar with the idea that communications hardware can create both operational dependence and security concerns. Optical transceivers are smaller and less visible, but they occupy a similarly sensitive position: they physically move data through the networks on which AI and cloud computing depend.
For American companies, the central risk is therefore not limited to a hypothetical cyberattack through one component. A concentrated Chinese supply base can also create economic leverage. If American AI firms become dependent on a narrow group of Chinese vendors, future export controls, sanctions, industrial policy changes or diplomatic conflict could disrupt data-center construction even without any technical compromise of the equipment itself.
This is where the issue connects directly to long-term U.S. AI competitiveness. America can lead in frontier models and advanced GPUs while still carrying hidden dependencies deeper in the infrastructure stack. A shortage of optical modules can delay new AI clusters just as effectively as a shortage of accelerators. Strategic resilience therefore depends on identifying bottlenecks before they become crises.
The proposed five-year transition reflects recognition that the United States cannot replace an entrenched supply chain overnight. Domestic manufacturers and trusted allied suppliers would need additional capacity, capital equipment, skilled labor and long-term purchase commitments before they could absorb demand now served by Chinese producers. Reuters reported that industry groups are already warning that abrupt restrictions could slow American data-center construction precisely when competition in AI is accelerating.
The most important lesson is that America’s AI infrastructure is only as independent as its weakest supply-chain dependency. The public debate often focuses on whose model is smartest or whose chip is fastest, but the physical network connecting those systems matters just as much. Optical transceivers are one of those overlooked components.
The new Senate bill has therefore exposed a broader reality: Chinese manufacturers are deeply embedded in parts of the infrastructure supporting America’s AI expansion, including equipment that can sit inside sensitive networks. Whether Congress ultimately adopts S. 5548 in its current form remains undecided, but the underlying supply-chain issue is already visible. The United States is entering an era in which AI security will depend not only on algorithms and semiconductors, but also on knowing who builds the hardware carrying the data between them.
