China dominates optical-module assembly, but much of the highest-value technology inside the modules remains concentrated in U.S. companies.
Co-packaged optics (CPO) could shift value away from the standalone transceiver box toward lasers, photonic components, and high-speed chips as optics move closer to the switch and, eventually, the AI accelerator.
Proposed U.S. restrictions on new Chinese-made transceivers could accelerate a transition already underway, reinforcing the value of owning the parts of the photonics stack where technology and IP are hardest to replicate.
In early August, Reuters reported that the FCC is drafting a rule to bar new Chinese-made optical transceivers from U.S. data centers.1 North American photonics stocks jumped. Understanding why, and what it means for China’s role in AI optical interconnects, matters for investors.
Seven of the world’s ten largest optical transceiver suppliers are Chinese. InnoLight alone accounts for roughly a quarter of global shipments,1 and Chinese vendors supply well over half the optical modules going into U.S. AI data centers.2
That number is real. It is also the least interesting number in the industry.
An optical transceiver converts electrical data into light, sends that data through fiber-optic cable, and then converts it back into an electrical signal. China is exceptionally good at making these boxes at enormous scale and at competitive prices. But the box is not necessarily where the most valuable technology sits.
What matters is what is inside the transceiver.
Two components are particularly important: The laser, which generates the light carrying the data, and the DSP, or digital signal processor, which processes the signals needed to transmit data reliably at extremely high speeds.
These technologies remain concentrated in American companies. Coherent and Lumentum are major suppliers of the lasers, while Broadcom and Marvell are major suppliers of the DSPs.
Chinese module makers combine these components with their own packaging and manufacturing expertise to assemble transceivers at enormous scale. They compete aggressively on cost, scale, and qualification speed, or how quickly products can be tested and approved by customers. It is a formidable business, but it is not the same as controlling the underlying technology.
And this is where the story gets more interesting.
Today, most optical transceivers are pluggable modules: Separate boxes that can be plugged into the front of a network switch. This makes them relatively easy to replace or upgrade. But the industry is moving toward something called co-packaged optics, or CPO.
CPO integrates the optical components directly onto the same chip package as the main networking chip, replacing the separate module. Future iterations of CPO could also see the optics packaged with the AI accelerator itself. CPO replaces more electrical connections with optical ones, even over short distances, improving bandwidth and power consumption compared with copper.
Traditional Pluggable Transceiver vs. Co-Packaged Optics
Source: Nvidia, Aug 2025
Broadcom’s CPO switches are shipping. NVIDIA’s Quantum-X switch reached commercial availability in early 2026. So what survives this transition and what does not? The laser stays. The silicon stays. The box gets absorbed into the chip package controlled by NVIDIA and Broadcom. That distinction matters.
The Chinese transceiver makers have built enormous businesses around designing, assembling, and shipping the box. If the industry increasingly stops using a separate box, some of that value will migrate elsewhere in the supply chain.
CPO is still early, representing approximately 0.5% of AI data-center optics today.4 This is a multi-year shift, not a switch that flips next quarter. But for investors, the direction of travel matters. As CPO takes share, the question becomes less about who can manufacture the most optical modules and more about who owns the critical technologies embedded inside the new architecture, particularly the lasers, photonic components, and high-speed chips.
The proposed FCC rule could accelerate this transition, prohibiting new Chinese transceiver models from entering U.S. data centers on national-security grounds, with officials reportedly wanting it enforced before year-end.
The logic is straightforward: If Washington sees Chinese networking equipment as a security or supply-chain risk, it is better to prevent it from becoming embedded in America’s AI infrastructure than to remove it later. Nothing has been published yet, and the proposal could still be modified or shelved. But even the possibility changes the investment equation.
Engineering was already moving the industry toward architectures in which the traditional transceiver becomes less important. Regulation could accelerate that process.
Our thesis is simple: Own the parts of the photonics supply chain where technology is hardest to replicate and value is likely to remain concentrated. That means lasers, manufacturing equipment, and chips, rather than simply the assembly of optical modules.
The Tema Photonics & Optical ETF (LAZR) selectively owns Chinese photonics leaders where we see durable competitive advantages, reflecting their important role in today’s supply chain. At the same time, we are less concentrated in module assembly, where the shift toward CPO and the prospect of additional U.S. restrictions create greater structural uncertainty.
That uncertainty is especially difficult to stay ahead of. Longstanding U.S.-China trade and technology tensions, combined with a policy environment that can change quickly under the current administration, mean the regulatory path is inherently unpredictable. For LAZR, that reinforces the importance of selectivity and expert-driven active management.