Recently, I noticed a particularly interesting phenomenon—the upgrade demand for AI data centers is giving rise to an entirely new industry chain. The topic of CPO concept stocks has also been gaining increasing popularity in the investment community, and it’s worth discussing in depth.

To put it simply, traditional copper-wire transmission has become a bottleneck. With data volumes exploding, copper wires not only generate excessive heat, but also can’t keep up with transmission speeds, resulting in especially high power consumption. As a result, the industry is starting to shift toward using light to transmit data—this is the core logic of silicon photonics: shrinking the originally bulky optical components (lasers, detectors, and modulators) down to chip level and integrating them onto a silicon-based substrate.

And CPO (Co-Packaged Optics) is one way to bring this concept to life. Simply put, instead of using traditional pluggable optical modules, optical modules are directly packaged next to the CPU/GPU, greatly reducing transmission distance and saving more than 30% of energy. Why are these two concepts so often linked? Because to realize CPO, optical components must become extremely small and very low-profile, and silicon photonics can do exactly that. In plain terms, silicon photonics is the core technology, while CPO is the most promising application outlet.

Entering 2026, this industry is officially entering the stage of large-scale mass production. The opportunities across the entire supply chain are especially significant—from the leading chip design companies in the US stock market (NVIDIA, Broadcom, Marvell), to Taiwan’s wafer foundries, advanced packaging and testing, and optical component manufacturers—forming a complete ecosystem.

In the US market, Broadcom is leading in the CPO space, and its Tomahawk series has become standard equipment for AI data centers. Marvell stands shoulder to shoulder with Broadcom in the field of optoelectronic conversion chips and has just announced deep cooperation with NVIDIA. Meanwhile, Microchip (Marvell) acquires DustPhotonics, directly gaining control of photonic integrated circuit technology, with complete solutions available from 800G to 1.6T. As leading optical component players, Coherent and Lumentum are also accelerating their shift toward silicon photonics solutions.

In Taiwan, the opportunity for CPO concept stocks is even more direct. TSMC is not only a chip foundry, but also helps define CPO packaging standards. Its COUPE platform is central to silicon photonics development, with mass production planned for 2026. Tongxin-KY and ASE Technology (ASE) are the main forces in advanced packaging and testing, and they hold leading technologies in 800G and 1.6T high-speed transmission modules. Accton (Ateam) and TSMC are working in close depth to develop fiber array interconnect technology—this “interface” is particularly critical. Greatek (Browave) controls optical passive components, Lianyu provides the laser light sources required for CPO, and AnChuan’s optical-path positioning and inspection technology is key to improving yield.

However, investing in CPO concept stocks also requires attention to several risks. First is the issue of yield. Since optical components are integrated with chip packaging, if just one component fails, the entire GPU could potentially be scrapped. When reviewing financial reports, pay special attention to the gross margin trend—if revenue increases but gross margin declines, it may indicate that yields are still struggling. Second is the battle over specifications. LPO (linear pluggable optical modules) may capture part of market share; if your target is a pure CPO shop, be mindful of competitive threats. Third, confirm whether the company’s optical communications revenue share is truly increasing significantly—don’t be misled by concept stocks that are only riding the hype. Finally, there are geopolitical risks: the US broadband infrastructure plans and export controls on advanced semiconductors will directly affect this industry.

Frankly, silicon photonics and CPO are not short-term themes, but structural growth trends over the next 5 to 10 years. With 2026 serving as the watershed between R&D validation and large-scale mass production, it will truly test each company’s ability to implement the technology. The investment logic is simple: US stocks focus on standard-setting, while Taiwan stocks focus on supply-chain execution. While chasing new themes, remember to return to fundamentals and prioritize companies that have passed major customer certifications and have a clearly visible increase in optical communications revenue share. That’s how you can avoid noise in this fast-moving race and seize genuinely valuable high-quality investment opportunities.