The U.S. Federal Communications Commission, or FCC, has formally published its final rule, titled Protecting Against National Security Threats to the Communications Supply Chain Through the Equipment Authorization Program, in the Federal Register process. The rule was approved on July 22, is scheduled for formal publication on Sept. 11, and will take effect 30 days later.
What many in the market had worried about most did not happen. Core Chinese companies in the optical communications supply chain, including Eoptolink, Innolight, Suzhou Dongshan Precision Manufacturing and TFC Communication, were not directly added to an FCC restriction list.
The concern did not come out of nowhere. On July 22, the FCC brought optical transceiver modules into the Cover List control scope, barring imports of equipment containing logical hardware components made by Chinese entities on the covered list. On Aug. 4, the market also saw reports of a draft proposal that would ban imports of new Chinese optical transceiver models. Now that the final rule has arrived, the most severe scenario of a comprehensive blockade has, for now, not become reality. The struggle around supply chain security and substitution, however, is still very much alive.
The shift is no longer only about a restriction list
The report argues that the optical module industry is facing a change deeper than a single blacklist decision. Co-packaged optics, or CPO, is approaching mass production, and companies such as NVIDIA and Broadcom are steering a new round of change in data-center optical interconnects.
NVIDIA’s CPO design, as described in the article, abandons the traditional pluggable optical module and instead uses silicon photonic devices packaged together with ASIC chips. Compared with the conventional approach, the technology can improve energy efficiency by 5x, significantly improve network reliability, and increase continuous business application uptime by 5x as well.
A hard cutoff would also hurt the U.S. side
Even though the final rule did not directly restrict Chinese companies in the way some rumors suggested, the episode has exposed a basic problem that cannot be ignored: U.S. AI infrastructure construction still depends heavily on China’s optical module supply chain, and a forced cutoff would bring losses on both sides.
From a global industry perspective, Chinese manufacturers hold a dominant position in optical modules. Seven of the world’s top 10 optical module vendors are Chinese companies. In high-speed segments such as 800G and 1.6T modules, combined shipments from Chinese suppliers account for more than 70% of the market. Eoptolink alone holds about 27% of the global data-center optical transceiver market, according to the article.
U.S. companies Coherent and Lumentum remain technologically competitive, but their production scale is far from enough to replace Chinese supply capacity. The report says industry participants broadly believe overseas optical component suppliers are generally unwilling to pursue businesses with gross margins below 50%, while the complete industrial chain and efficient delivery capabilities in China and the broader Asia-Pacific region remain hard to replace in the short term.
The cost of any forced ban would be borne directly by downstream industries in the United States. On one side, procurement costs for U.S. cloud service providers such as Amazon Web Services, or AWS, would rise sharply. On the other, in a period when AI infrastructure buildout is accelerating, any supply gap would directly slow data-center construction and deployment. An analysis cited from Bank of America says a broad import ban would not only disrupt existing supply-chain order but could also delay the rollout of next-generation data-center architectures, running against the U.S. goal of speeding up AI computing infrastructure.
Dependence runs in both directions. Core components upstream of optical modules, including DSP chips and analog chips, are supplied in many cases by U.S. companies such as Marvell and Broadcom. A sweeping blockade aimed at Chinese modules would eventually hit revenue at upstream U.S. chip suppliers too. The article notes that since 2020, the market has repeatedly worried that Chinese optical module makers would lose share to North American rivals. Yet after years of testing that assumption, Chinese companies have still increased, not lost, global market share despite continued U.S. industrial policies aimed at supporting domestic supply chains.
Chinese vendors have been preparing for years
The article says Chinese optical module companies have not been standing still. The industry began preparing for geopolitical risks five or six years ago.
Capacity built outside China
The first response has been global manufacturing expansion to offset market-access risk. Leading suppliers such as Eoptolink began expanding capacity in Thailand and Malaysia in 2024. By 2025, their factories outside China had gained the ability to meet U.S. market demand independently, reducing exposure to rules tied to country of origin.
At the same time, major North American cloud companies have already carried out deep joint development with Chinese firms on next-generation products such as NPO and Coherent lite. The report says those development cycles and technical barriers cannot be replaced quickly, which has further tightened the commercial relationship between the two sides.
Multiple bets on next-generation architectures
The second response has been to spread bets across multiple technology paths instead of relying on one track. Current restrictions are still mainly aimed at pluggable modules, but Chinese suppliers are not limiting themselves to that category.
In near-packaged optics, or NPO, the report says Chinese companies already hold a visible lead. In April 2026, Google placed an order for 12 million NPO optical modules for its next-generation TPU v7, v8 and v9 supercomputing clusters. Eoptolink and Innolight won 60% and 40% of the order, respectively, taking the whole allocation.
On the CPO side, companies including Eoptolink and Innolight have clearly defined a shift from selling modules to selling optical engines. Eoptolink has already built reserves in silicon photonic chip technology and is testing a 3.2T optical engine, with the goal of becoming an indispensable supplier of optical components in the CPO ecosystem. Innolight has launched 1.6T optical module products spanning VCSEL/EML, silicon photonics and thin-film lithium niobate routes, while also laying out high-speed LPO and CPO-related products.
Overseas joint ventures as a route into the high end of the market
The third response has been a change in business model through overseas joint ventures designed to break through geopolitical barriers and enter the global high-end supply chain. The report presents TFC Communication’s project with SuperX as a representative case.
On April 21, 2026, TFC Communication’s wholly owned subsidiary in Singapore, together with Nasdaq-listed SuperX and APEX VERVE LIMITED, signed an agreement to contribute SGD 2 million to establish SuperX Optical Communications Pte. Ltd. in Singapore. SuperX holds 45%, Singapore TFC holds 35%, and APEX VERVE LIMITED holds 20%. The joint venture completed registration and full capital contribution on July 1, 2026.
The article lays out three functions for this setup. First, by registering the company in Singapore and covering global markets outside mainland China and Hong Kong and Macao, the venture can bypass FCC covered-list restrictions aimed at mainland Chinese entities. Second, it helps TFC Communication tie itself more closely to the CPO core ecosystem. At NVIDIA’s GTC 2025 conference, TFC Communication was publicly listed as a core partner for the Quantum-X and Spectrum-X silicon photonics CPO switches, alongside companies including TSMC, Coherent and Corning. Third, it supports industrial upgrading by moving TFC Communication from being an upstream passive-component supplier into direct participation in worldwide sales of silicon photonic modules and optical engines.
China also has leverage in upstream materials
Beyond company-level moves, the report says China holds a natural balancing position in upstream raw materials used for optical chips.
Indium phosphide, or InP, is widely used in DFB lasers, EML lasers and photodetectors, making it a necessary material for 800G, 1.6T and next-generation 3.2T optical modules. According to the article, production capacity for the high-purity indium metal required for InP optical chips is largely concentrated in China. Even U.S. company AXT has located its InP factory inside China, and its subsidiary Beijing Tongmei accounts for about 36% of the global InP substrate market.
The market is also dealing with a severe supply-demand mismatch. As optical module speeds move toward 1.6T and 3.2T, demand for InP substrates from AI data centers is expanding sharply. The article says the global gap between supply and demand for InP substrates had already exceeded 70% in 2025, and that the tightness is expected to continue through 2027. Prices have responded. The average price of a 4-inch substrate has risen about 50% from the start of the year, while quotations for refined indium have jumped by more than 80%. NVIDIA has already asked suppliers to expand InP laser production capacity to 20x the current level by 2030.
In January 2026, China’s Ministry of Commerce announced a full ban on exports of dual-use items to Japanese military end users and uses, including InP, indium, gallium and germanium. Civil exports remain subject to strict licensing and end-user review. In the article’s framing, that creates a reverse balancing mechanism in the broader industrial contest: if the United States were to ban imports of Chinese optical modules, China could respond by tightening export controls over key upstream materials. Without China’s indium supply, production at U.S. domestic optical-chip manufacturers would also stall, reducing the practical feasibility of any extreme ban.
The deeper pressure comes from technology transition
In the article’s view, the impact of external restrictions is limited compared with the challenge of technological turnover. In August 2026, Gilad Shainer, senior vice president of networking at NVIDIA, formally stated that CPO had entered mass production. Spectrum-X CPO switches have already been delivered to core customers and deployed inside NVIDIA’s own AI factories.
For now, expected controls are concentrated on pluggable optical modules, which is exactly the category where Chinese vendors are strongest. In the short term, pluggable modules are not likely to be fully replaced by CPO. Over the medium to long term, though, the direction of travel is becoming clear. The report says that in the CPO era, command over the supply chain will sit with companies that can manufacture across the full optical-electrical integration chain, and this remains an area where Chinese industry still has a visible weakness.
Overseas silicon photonics foundries are already in position
Looking abroad, TSMC is described as a leader in CPO through its COUPE platform. The report says the company is using its advanced-process capabilities to build high-end photonic integrated circuit manufacturing, focusing on 3D stacking and optoelectronic co-packaging. It is closely tied to top-tier customers such as NVIDIA and Broadcom and serves as a key source of high-end silicon photonics foundry capacity for next-generation CPO architectures. Its photonic integrated circuit capacity is climbing along with AI compute demand.
GlobalFoundries, after acquiring Singapore-based AMF, has become one of the largest pure-play silicon photonics foundries by scale. Its Fotonix platform focuses on data-center and AI optical interconnects. According to the article, it has manufacturing capacity in both the United States and Singapore, more than 10 years of silicon photonics work, and coverage across both pluggable transceivers and CPO devices. In 2026, it received $300 million in U.S. government funding for next-generation silicon photonics research, with a target of raising single-channel speed to 400Gb/s.
UMC is concentrating on mature-process silicon photonics. The report says the company has allied with IMEC, with UMC handling the front-end wafer manufacturing process on a 28-22nm platform. Trial production is expected in 2026, followed by volume output in 2027. On July 14, 2026, UMC and Singapore photonic chip company SILITH Technology jointly announced that UMC’s 12-inch Fab 12i in Singapore had completed delivery of its first batch of volume-produced silicon photonics wafers, moving their joint 1.6T silicon photonics platform from development into commercial mass production.
Tower Semiconductor, headquartered in Israel with fabs in the United States and Japan, is another name highlighted in the report. Unlike foundry giants focused on advanced logic nodes, Tower concentrates on high-value analog and mixed-signal manufacturing, with differentiation in specialty processes such as silicon photonics and silicon-germanium. In February 2026, Tower announced a partnership with NVIDIA to support its 1.6T data-center optical modules with a silicon photonics platform. The article treats that as a sign that Tower’s capabilities in AI optical interconnects have won recognition from NVIDIA.
Against that backdrop, the report says Chinese domestic companies still lag in silicon photonics foundry capability.
Holding market share is one task; catching up in process technology is another
Overall, the article concludes that Chinese optical module makers are operating at a point where geopolitical pressure and next-generation technology change are colliding. On Sept. 11, 2026, the FCC’s final rule on the relevant equipment formally landed, but it did not directly restrict Chinese companies as rumored and did not create a comprehensive ban on Chinese optical modules. External pressure has not, at least for now, turned into the harshest possible outcome. Leading Chinese companies have also been working to offset geopolitical risks through overseas capacity shifts, diversified next-generation technology bets and offshore joint ventures, helping preserve their strength in the global high-speed pluggable module market.
But the article’s central point is that external risk is not the industry’s biggest threat. As CPO heads into mass production, the optical interconnect sector is moving into a new lane. Future competition will center on full-chain optical-electrical integration capability. TSMC, GlobalFoundries and UMC have already built technology and customer reserves in silicon photonics foundry work. By comparison, silicon photonics manufacturing remains a notable weak point for Chinese players. For domestic optical module companies, scale advantages in traditional pluggable modules alone may not be enough to last. Keeping their current global foothold while closing gaps in process technologies such as silicon photonics foundry, and shifting from module vendors to providers of optical engines and integrated optoelectronic solutions, is presented as the key task in the next era of AI optical interconnects.
The article was originally published by the WeChat public account Semiconductor Industry Vertical, ID ICVIEWS, and credited to the ICVIEWS editorial team.