Chinese optical transceiver ban: FCC moves ahead despite lobbying

Share
Chinese optical transceiver ban: FCC moves ahead despite lobbying

The most consequential AI rule of the week concerns a component nobody photographs: the small pluggable module that turns electricity into light between racks. Washington is drafting a ban on new Chinese models, and the companies that buy them are losing the argument.

The Trump administration is pressing ahead with a plan to bar imports of new Chinese optical transceivers, despite pushback from Amazon, Google, Meta and Microsoft, Reuters reported on Tuesday. The measure is being drafted at the Federal Communications Commission, which oversees the equipment authorizations every network device needs before it can be imported and sold in the United States. It has not been published. The agency can still amend it, narrow it, or drop it, and officials have been aiming to issue it before the end of the year, Reuters reported when the drafting became public in August.

What the rule would do is narrower than the headlines suggest, and the narrowness is the interesting part. It targets new models — transceiver designs that have not yet received FCC authorization — rather than the hardware already installed and certified. Equipment sitting in a rack in Ohio stays legal. The next generation of modules, the 1.6-terabit parts that the biggest clusters are just beginning to order, would arrive without a paperwork path if they come from a Chinese vendor.

Optical transceivers are the plumbing of an AI cluster, and the plumbing is now a bottleneck. Above a few metres, copper cannot carry 200 gigabits per lane without the signal falling apart, so every link between GPU servers, switches and racks runs over fibre, and every end of every fibre needs a module that converts electrical signals to light and back. A single 576-GPU GB200 system runs tens of thousands of 800-gigabit modules, by the industry's own counts. Scale a cluster to 100,000 accelerators and the optical layer is no longer a line item; it is the thing that decides whether the compute you bought can talk to itself.

China owns most of that layer, which is why the ban exists. LightCounting's vendor rankings put seven Chinese companies among the world's top ten optical module suppliers, with a combined share above 60 percent, as the Global Times summarised. Counterpoint Research credits Zhongji Innolight alone with roughly 27 percent of the global data-center transceiver market. Innolight and Chengdu-based Eoptolink supply the majority of Nvidia's 800G demand, price 20 to 25 percent below Western incumbents, and earn most of their revenue outside China — Innolight's customer list reportedly runs Alphabet at about 22 percent of revenue, Amazon at 11 percent and Meta at 6.4 percent. Innolight was added in June to the Pentagon's list of companies it alleges are backed by the Chinese military, an designation that has usually preceded tighter restrictions rather than replaced them.

The intended beneficiaries are the American optics houses: Coherent, Lumentum and Applied Optoelectronics, plus Marvell, which is mid-way through buying Lumentum. They rallied hard when the draft leaked in August — Coherent up 11 percent, Applied Optoelectronics 18 percent, Lumentum 7 percent on the day, while Innolight's Shenzhen shares fell as much as 14 percent. Lumentum's own numbers show why the market treats a policy headline as an earnings event: its revenue has been doubling as data centers go optical, a shift we looked at in Lumentum's revenue doubles as AI data centers go optical.

The stated rationale is security, and it is not frivolous. A modern transceiver is not a dumb glass-and-laser part: it carries firmware, onboard memory and a management interface that the network can query and update. That is enough surface for compromised firmware, manipulated diagnostics, or a module that misbehaves at a chosen moment, and China's 2017 National Intelligence Law obliges domestic organisations to assist state intelligence work regardless of what a vendor's privacy policy says. What the record does not contain is a confirmed incident: no publicly documented case of a Chinese transceiver being used to exfiltrate data from a US facility. The risk is architectural and legal, not evidenced. The FCC has built a pattern on that basis anyway — Huawei in telecom gear, DJI in drones, Chinese-made consumer routers in March, humanoid robots and power inverters in July.

Then there is the part the lobbying campaign is right about, and it is bigger than the cost argument Microsoft has been making. The Western suppliers the ban is meant to empower build their lasers on indium phosphide, and China refines roughly 70 percent of the world's indium. Beijing put indium phosphide on its export control list in February 2025; since then the price of a six-inch indium phosphide wafer has risen about 250 percent, from roughly $1,400 to roughly $5,000, and AXT, a key substrate supplier to Coherent, has carried a backlog above $60 million. Lumentum's chief executive, Michael Hurlston, has said the indium phosphide shortage will be worse than the memory crunch. American 800G assembly capacity sits at roughly 100,000 units a month against demand that LightCounting already puts about 30 percent above supply, as Tech Times laid out. In August the Information Technology Industry Council told the FCC the restriction could grind data-center rollout to a halt, a warning Tom's Hardware reported alongside Microsoft's own lobbying.

Read the scope again and the policy's real shape appears. A ban on new models does not remove Chinese optics from American data centers; it freezes the generation that is already certified and makes the next one a licensing question. That converts a price-competitive commodity market into an approval-queue market, and in an approval queue incumbency is the asset — which is a fine outcome for the Western vendors, a survivable one for hyperscalers still deploying 800G, and a hard stop for anyone who wanted 1.6T from the cheapest supplier on schedule. The firms lobbying against it are not arguing about security. They are arguing about who absorbs a two-year capacity gap.

The deeper problem is that the substitute supply chain runs through the country the rule targets. If Beijing answers a transceiver ban by slowing indium phosphide permits — a lever it has already pulled once, without a ban to retaliate against — the Western ramp the FCC is counting on cannot arrive on time no matter how the exemption list is drafted. That is the argument worth making to the commission, and the industry is making a weaker one about procurement costs.

What to watch: whether the final text covers only new authorizations or reaches deeper into installed gear; how the FCC defines a "Chinese" transceiver once Innolight and Eoptolink are assembling in Thailand, a definition Bloomberg Intelligence's Sean Chen has flagged as the difference between a real ban and a paperwork one; whether non-Chinese suppliers get conditional approval fast enough to matter; and whether Beijing reaches for indium permits rather than tariffs. The FCC has until the end of the year, and Nvidia reports earnings tonight into a supply chain that just got a new rule.

Would you pay 20 percent more for optics with a US paperwork trail, or route around the rule through a Malaysian colo? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Pax Silica allies agree a joint list of AI tech to shield from China

Pax Silica allies agree a joint list of AI tech to shield from China

The coalition spent a year signing declarations. On Sunday it agreed to start naming the things it will actually defend. The United States and eight allies agreed to build a joint list of AI technologies and supply chains that need protecting from China, signing the San Francisco Declaration on the Secure Foundations of Enduring Prosperity, the Financial Times reported. Japan, Australia, Israel and the United Kingdom signed, and South Korea, India, Qatar and the United Arab Emirates did too.

China AI export rules: DeepSeek and MiniMax clear registration

China AI export rules: DeepSeek and MiniMax clear registration

Two of China's biggest open-weight labs cleared Beijing's new AI export registration in under a week. In Seoul the same morning, the memory champion found out what the capital markets will pay for an AI-adjacent listing right now — nothing. DeepSeek and MiniMax have completed the export registrations that Beijing's draft AI export-control rules require, and have gone ahead with open-source releases and overseas licensing, according to Chinese state media.

Renting US AI chips from abroad now needs a license

Renting US AI chips from abroad now needs a license

The Commerce Department published an interim final rule on Monday that turns renting American compute from a data center in a third country into a licensable act whenever the customer is headquartered in China or Macau. It closes the route the Megaspeed affair put on every front page last week, and it landed the day before Nvidia reports earnings. The mechanics are worth reading closely, because this is the first time Washington has regulated compute by the hour rather than by the…

Deep Dive — Can Jensen Huang return to Nvidia? 145,000 say yes

Deep Dive — Can Jensen Huang return to Nvidia? 145,000 say yes

By Sunday morning the "Stand with Jensen" open letter carried 145,549 signatures, according to the campaign's own counter — and 7,224 of them self-identified as current Nvidia employees, which the site puts at 17.2% of a workforce of roughly 42,000. The breakdown under that number is the part a board actually reads: three senior vice presidents, 23 vice presidents, 43 senior directors, 110 directors, 245 managers, and 2,913 individual contributors who gave a level. The letter's demand is…

Moonshot is shopping for 8,000 Nvidia Blackwell GPUs in Europe

Moonshot is shopping for 8,000 Nvidia Blackwell GPUs in Europe

Moonshot AI has spent this year trying to rent its way onto the silicon Washington will not let it buy. The Beijing lab behind Kimi talked to neocloud operators about leasing Nvidia Blackwell capacity hosted in European data centres, according to four people familiar with the matter who spoke to CNBC — clusters of 8,000 or more GPUs, three of them said. Two described the interest as inference capacity, the compute that serves finished models to users; one also knew of interest in training.

Data center bans swept the ballot — Texas flipped, 13 GW at risk

Data center bans swept the ballot — Texas flipped, 13 GW at risk

For two years the fight over AI infrastructure was fought in planning commissions and interconnection queues. On Tuesday it was fought at the ballot box, and the industry lost nearly every vote it was in. Voters restricted data center projects in almost every local measure put before them, and Texas — the state expected to host a quarter of the American data center capacity coming online by 2030 — elected a Democratic governor who campaigned on halting new builds. 404 Media's Jason Koebler…