The next front in the US and China technology contest is not a chip or a phone. It is a part about the size of a lighter that most executives have never heard of. On 10 November Reuters reported, citing three people familiar with the matter, that Microsoft and several industry groups are pressing the Trump administration to abandon a planned ban on imports of new Chinese optical transceivers.
Transceivers convert electrical signals into light and back again, so that data can race over fibre between servers, switches and the thousands of accelerators in an AI cluster. Every large data centre uses them by the tens of thousands. Microsoft declined to comment on the report.
What Washington Is Weighing
The Federal Communications Commission has been drafting the restriction since at least August, when Reuters first reported that it would block imports of new models of Chinese-made transceivers. The stated concern is that Beijing could use the devices to steal data, plant malware or disrupt service at American facilities.
The scope is still moving. Two sources told Reuters the FCC is considering a threshold of 3.2 terabits per second, which would catch the next generation of high-speed modules rather than the 800 gigabit and 1.6 terabit parts shipping in volume today. Morgan Stanley told clients in October that any rule would probably phase in starting with 3.2 terabit products, which are not expected to ramp until around 2028 and 2029.
Congress is moving on a narrower track. In late September senators Dave McCormick, Ruben Gallego, John Cornyn and John Fetterman introduced a bipartisan bill to bar Chinese transceivers from national security systems. It names Zhongji Innolight and Eoptolink, allows a five-year transition and asks the Commerce Department to assess US and allied production capacity. Innolight was added earlier this year to the Pentagon’s list of Chinese military companies operating in the United States.
Why Microsoft Is Worried
The problem is concentration. Chinese manufacturers, led by Innolight and Eoptolink, account for roughly two-thirds of global transceiver sales, according to Counterpoint Research. Western suppliers such as Coherent and Lumentum exist, and their shares jumped by 11% and 7% respectively when the plan first surfaced in August, but Counterpoint estimates it would take them 12 to 24 months to build the cleanroom, packaging and yield capacity to fill the gap.
Microsoft has more riding on that timeline than most. Data from Cleanview cited by Reuters shows it has 35 data centres under construction in the US and another 134 planned. A supply squeeze on a component it buys in enormous quantities would add cost and delay to that pipeline at the very moment it is racing Amazon, Google and Oracle for AI capacity.
The Information Technology Industry Council, which wrote to the FCC on 13 August asking it not to add transceivers to its restricted list, has been blunt. Its general counsel, John Miller, said the ban “really threatens to arguably grind” the data centre rollout to a halt.
The timing is awkward for Washington too. Beijing and the US have just extended their trade truce into January, with a Trump and Xi meeting expected at APEC later this month. A fresh hardware restriction could become a bargaining chip, or a reason for China to tighten its own controls on materials such as the indium used in optical components.
What the Optics Fight Means for Your Business
The impact. Most companies will never buy a transceiver directly, but they rent them every time they pay a cloud bill or sign a colocation contract. A ban, even one limited to 3.2 terabit parts, would push up the cost and stretch the lead times of the next wave of AI capacity, and hyperscalers will pass part of that on. The most exposed are firms planning large AI training or inference projects for 2027 and 2028, operators building private data centres, telecom and network equipment buyers, and any business whose growth plan assumes cloud prices keep falling. Suppliers of Western optics, and the investors behind them, stand to gain.
What to do now:
1. Ask your cloud and colocation providers in writing how much of their optical supply comes from Chinese vendors, and what their plan is if new models are barred.
2. Review multi-year cloud and AI capacity contracts for clauses that let providers pass through component or regulatory costs, and negotiate caps before renewals.
3. If you run your own data centres or networks, audit installed and planned transceivers by origin and speed, and qualify at least one non-Chinese supplier now while lead times are manageable.
4. Build a scenario into 2027 budgets where AI compute costs rise rather than fall, and decide which projects still pay back under it.
5. Brief the board on the wider pattern: security rules are reaching ever deeper into the supply chain, and the next one may hit a component your business does buy.
Whether Microsoft wins this round or not, the direction is clear. The hardware inside the cloud is now a matter of national security, and its price will be set partly in Washington and Beijing.
