U.S. Mulls Ban on Chinese Optical Modules, AI Data Centers Face Supply Gap
0xBroomberg
The FCC is reportedly studying restrictions on Chinese-made optical transceivers, yet Chinese vendors account for roughly two-thirds of global shipments — a ban would stall AI data-center buildouts before western suppliers can fill the gap.
What are optical transceivers, and why do they bottleneck AI?
Optical transceivers — small devices that convert electrical signals into light — are the nervous system linking servers and switches inside a data center. Without them, GPUs cannot talk to each other.
AI clusters are migrating from 400G to 800G and 1.6T optical networks, driving demand and technical requirements sharply higher.
This means → the ban targets not an optional accessory but a physical chokepoint for the entire AI infrastructure stack.
How dominant are Chinese suppliers?
Counterpoint Research estimates Innolight (中际旭创) leads global data-center transceiver revenue at roughly 27%; Coherent holds about 17%.
Innolight, Eoptolink (华工正源), Accelink (武汉光迅), and Source Photonics together account for roughly two-thirds of global shipments and about 60% of datacom transceiver revenue.
In plain terms = two out of every three optical transceivers shipped worldwide come from China. That gap cannot be closed by tightening supply — it requires building entirely new capacity.
Why can't western suppliers fill the gap quickly?
Coherent, Lumentum, and other western vendors compete on photonics technology, but lack the cleanroom capacity, automated packaging lines, and high-yield volume production to absorb Chinese output in the near term.
Transceivers are not commodity parts. Each module must pass compatibility testing with switches, DSPs — digital signal processors — and the broader data-center architecture. Switching to a new qualified supplier triggers a lengthy re-certification cycle.
This means → the bottleneck is not just "can they build it" but "will it work in our racks." Analysts see this gap persisting for one to two years.
What does this mean for hyperscalers versus smaller operators?
At 800G and 1.6T speeds, a shortage of a relatively cheap transceiver can leave far more expensive GPUs sitting idle, delaying multi-billion-dollar AI cluster launches.
AWS, Microsoft, Google, and Meta have stronger bargaining power and may secure replacement supply first.
This reflects a harsher reality: smaller data-center operators and colocation providers will bear the brunt of price increases and delivery delays, widening the industry divide.
How the ban is worded will determine how hard it hits — what are the scenarios?
Reports suggest restrictions may target new models only, preserving the installed base but constraining next-generation procurement.
If the rule keys on parent-company ownership rather than manufacturing location, the blast radius widens sharply — Innolight and Eoptolink already run automated lines in Thailand, and an ownership-based rule would sweep in Southeast Asian output too.
In plain terms = the same ban, worded differently, could vary the impact by multiples. The pivotal question: "Are we banning what China makes, or what Chinese companies make?"
Who wins, who loses — and why isn't this a simple "US wins, China loses" story?
If restrictions persist, Coherent, Lumentum, and Applied Optoelectronics stand to gain market share and pricing power. Cisco's Acacia unit and suppliers in Japan and Taiwan could also benefit from supply-chain diversification.
But Chinese module makers rely heavily on Broadcom and Marvell DSPs, and on lasers and optical components from Lumentum, Coherent, and Mitsubishi Electric — the ban simultaneously hits US upstream firms selling into China.
This reflects the real structure of the supply chain: it is not two separate ecosystems facing off but one interdependent chain being severed in the middle — both sides bleed.
Content is for reference only, not financial advice.