Reuters reported this morning that the FCC is preparing a measure that will ban U.S. imports of new Chinese optical transceiver modules. Officials aim to publish it this year. What constitutes “new”? What constitutes “Chinese”? Not clear, and we explore this later.
To get ahead of the questions – and to state plainly that Cignal AI does not expect this measure to stick – here is what we think:
Who is exposed?
The most exposed supplier would be Innolight, to which Cignal AI attributes $2.6B of datacom optical component revenue in 1Q26, or 34% of the $7.7B worldwide market. Following are Eoptolink, HG Genuine, and possibly Source Photonics. As a group, Chinese-headquartered suppliers account for roughly 60% of all datacom optical component revenue.
Where is Reuters wrong?
Cignal AI’s measurement of Innolight’s share is higher than the figure Reuters cites. Innolight’s 34% share of 1Q26 datacom component revenue is up from 25% a year earlier, based on revenue doubling (see the Cignal AI 1Q26 Optical Component Report and Datacom Optical Component Revenue Doubles to $7.7 Billion in 1Q26).
The three Chinese suppliers a ban would most directly impact – Innolight, Eoptolink, and HG Genuine – hold 51% of the market between them. Source Photonics, which manufactures in China, adds a further single-digit share.
How significant is Innolight in the 800G and 1.6T transceiver markets?
Innolight is the largest manufacturer of optical modules in the world. Cignal AI estimates that over 60% of its business is with U.S. hyperscalers. At $2.6B in 1Q26, Innolight is roughly 2.4 times bigger than the next largest supplier, Eoptolink at $1.1B.
Could Broadcom, Coherent, Lumentum, Applied Optoelectronics, and other non-Chinese suppliers replace Innolight’s volume?
Not anytime soon. Together, these companies supply most of the high-speed EMLs and assemblies as well as high-performance light sources, but they don’t have the manufacturing throughput to fill the void.
Applied Optoelectronics is small by comparison; $81M of datacom component revenue in 1Q26 against Innolight’s $2.6B. Coherent has manufacturing capabilities, but some are in China. Lumentum manufactures outside China, but it exited the module business in the past and only recently re-entered via its Cloud Light acquisition.
How quickly could alternative suppliers expand production?
Fabrinet and other contract manufacturers could expand, but the industry is already supply constrained. Fabrinet said as much on its most recent call, indicating widespread shortages held shipments below demand (1Q26 Vendor Reports and Key Takeaways). It would be years before Fabrinet could ramp and replace Innolight, Eoptolink, and HG Genuine.
Which part of the supply chain would be hardest to scale?
Manufacturing the modules. Most of the lasers at the high end are made outside of China, so the binding constraint is assembly, test, and qualified capacity rather than the optical chips. There is already a supply crunch for lasers, but capacity additions are underway.
Does this impact coherent optical transceivers?
No, not to any great extent. Marvell, Acacia, and Ciena dominate the 400G generation, while Nokia is added to the list for 800G. These are all non-Chinese companies with module manufacturing based outside of China. The blast radius is within Datacom transceivers, and if it is just “new” designs, it will be primarily 400G and above as well as emerging ELSFPs for CPO/NPO assembled in China.
Could hyperscalers switch suppliers without redesigning networks or delaying deployments?
Hyperscalers cannot switch suppliers easily, but that is not the real problem. The real problem is that alternative suppliers do not have the ex-China manufacturing capacity.
Would a restriction covering only “new modules” materially affect Innolight or the wider market?
Certainly. Innolight is positioned to be a leading maker of 3.2T modules, which enter production in 2028, according to Cignal AI’s forecast. Innolight has also publicly stated that it is participating in Google’s 2.4T Coherent Lite project. It’s unclear what “new” means – revisions and cost reductions to 800G and 1.6T optics would certainly be considered new. The 1.6T generation is still in early stages and will have many revisions and supply chain adjustments. And there is another round of cost reductions coming at 800G (see 2Q26 Vendor Report MaxLinear raises DSP guidance but slips 1.6T production).
Does the restriction follow the company or the country of manufacture?
Unknown. Innolight and other Chinese vendors have rapidly been adding capacity in Thailand and Indonesia. Innolight is also using the new brand TeraHop for its non-Chinese business. Is it considered a Chinese module when sold by TeraHop with a Marvell DSP, Lumentum optics, and assembled in Malaysia? What if Cisco re-labels an Innolight-manufactured module?
Any rule tied to country of origin is straightforward to route around. An entity-based Covered List designation follows the company wherever it builds. Reuters seems to hint that the measure will be company-based, not manufacturing-origin-based.
What does retaliation look like in this supply chain?
Beijing has said it will respond. The Chinese embassy was quoted by Reuters as saying: “China will take all necessary measures in response to any action that causes material harm to its interests.”
High-end lasers are largely made outside China, but packaging, substrates, and passive components are not. China’s rare-earth export controls last year are a strong card, and they could effectively choke off InP-based processes and access to critical subcomponents like optical isolators. The U.S. has limited leverage and can’t respond if China withholds key components – it’s mutually assured destruction.
Is this comparable to the Huawei ban?
Reuters states “China hawks in the administration are keen to avoid another situation like Huawei, where telecom equipment made by the heavily sanctioned Chinese firm was so deeply embedded in U.S.”
This is a spurious argument, as transceivers (unlike the equipment provided by Huawei) do not have any external management connection. At their most basic, transceivers merely convert electrical signals to optical signals and back again. There is no storage or analysis capability, particularly given the speed at which the data is traveling. Management of the transceiver is via the (non-Chinese) equipment that the transceiver is inserted into. In many cases, hyperscalers specify the use of MACsec to encrypt the physical layer anyway. There is no bona fide security risk.
This situation is also unlike the Huawei ban in the breadth of its scope. The reality is that only about $150M of Huawei equipment was ever deployed in the U.S. In contrast, the optical transceivers potentially covered by this ban are deployed by all hyperscalers and many neoscalers, enterprises, and telcos. The impact is massive – roughly half of the world’s optical module manufacturing capacity is in scope.
What is the likely effect on prices, lead times, and the U.S. AI datacenter buildout?
Prices and lead times will go up, and the buildout will slow. The real question is whether this ban happens at all. The Trump administration explicitly excluded optical transceiver modules when tariffs were put in place last year, and Reuters notes that Commerce shelved a set of import restrictions including one targeting Chinese datacenter equipment after the October trade truce. So why ban them now? Cignal AI is not yet convinced that this situation is real.
Cui Bono? Who benefits?
Hyperscalers source most optics from Chinese manufacturers. They certainly won’t welcome such a massive disruption. U.S. companies like Lumentum, Broadcom, and Coherent already own the high ground selling the highest-margin components. They don’t want to be in the module manufacturing business, and companies like Nvidia and others already lean heavily on Fabrinet for manufacturing. Optical contract manufacturers like Fabrinet, Sanmina, and Jabil stand to benefit – but only with massive capacity investment and the confidence that the rules will not change again in six months if Washington reverses course. So far, experience has proved otherwise.
Therefore, in the short term, no one benefits from a ban on Chinese optical transceiver modules. The additional uncertainty just makes navigating what is already a supply-constrained market undergoing price inflation even more challenging.
Does this change Cignal AI’s forecast?
No. This is all very chaotic right now. We’re not changing anything until we believe this will happen.