CITIC SECURITIES released a research report stating that the next phase for AI computing clusters is shifting from "stacking computing power" to "competing on network efficiency." Laser chips are the core components of optical modules, determining their electro-optical conversion efficiency and generational product capabilities. The laser chip industry is set to benefit from the expansion of Scale-out/Scale-up network super nodes and the increased value brought by technological advancements, presenting significant growth opportunities. Furthermore, due to strong market demand and short-term rigid production capacity, the laser chip market is experiencing rapid growth alongside a structural supply-demand imbalance. The advantages of leading manufacturers are becoming increasingly prominent. The report expresses optimism regarding the future development potential of leading laser chip companies.

The core viewpoints of CITIC SECURITIES are as follows:

Laser chips constitute the core value of optical modules, with CW and EML being the dominant solutions for high speeds. Laser chips are light source chips based on III-V semiconductor materials that achieve stimulated emission; their core function is converting high-speed electrical signals into light signals of specific wavelengths. In the AI data communications market, laser chips are primarily categorized into VCSEL, EML, and silicon photonics solutions. The core drivers consistently revolve around two main themes: "increasing single-channel data rate" and "extending transmission distance." In the 100G and earlier eras, data centers primarily focused on short-distance interconnects within 100 meters, where VCSEL dominated due to its low power consumption and cost. Entering the 400G/800G era, single-channel rates jumped above 100G. VCSEL, limited by multi-transverse mode effects, struggled to surpass 30GHz bandwidth in mass production. Simultaneously, AI cluster interconnect distances extended to 500m-2km, highlighting the constraints of multimode fiber dispersion, leading to the phasing out of VCSEL as the primary solution. EML, integrating DFB and EAM (electro-absorption modulator) monolithically, balances high bandwidth and low chirp, quickly becoming the dominant solution for 800G/1.6T. In the 1.6T era, silicon photonics leverages mature CMOS technology to integrate modulators, detectors, etc., onto a single PIC chip, significantly reducing packaging complexity and power consumption. Furthermore, CW lasers can achieve 1-to-2 or 1-to-4 multiplexing through beam splitting, directly diluting overall costs and making this approach mainstream. The 1.6T transition is a key inflection point where silicon photonics shifts from an "option" to a "core solution." Within this scheme, the CW laser, as a mandatory component, is seeing its strategic value elevate from a "supporting part" to a "core bottleneck component."

AI is driving phenomenal growth in the high-speed laser chip market, with a strong trend of increasing volume and price. On the demand side, the need for AI large model training and inference continues to climb. Global cloud providers have significantly revised their capital expenditures upwards, and AI computing clusters are expanding relentlessly. Interconnect efficiency has become the core bottleneck for cluster computing. As cluster scales grow from tens of thousands to hundreds of thousands or even millions of GPUs, the optical module-to-GPU ratio increases substantially from 1:2.5 to 1:5 or even 1:12. The demand for optical modules is膨胀ing at a pace far exceeding the growth in GPU numbers, correspondingly driving a significant increase in demand for laser chips. Concurrently, optical interconnects are accelerating their penetration into scenarios traditionally served by copper connections, further opening up incremental market space. On the supply side, the expansion cycle for laser chip production is lengthy, typically 18-24 months. Core MOCVD equipment is currently monopolized by two American and German manufacturers, creating significant short-term rigidity in production capacity. Leading laser chip manufacturers have their capacity booked until 2027, with order visibility for some customers extending to 2028. According to CITIC SECURITIES estimates, the market size for high-speed laser chips (100G and above) is expected to surge from $930 million in 2025 to approximately $4.06 billion in 2026, a year-on-year increase of over 338%. It is projected to further expand to about $7.10 billion in 2027, representing a cumulative growth of over 7 times in two years. Within this, the supply gap for 200G EML chips is as high as 20%, and for 400mW CW lasers, it reaches 17%. This seller's market structure is unlikely to be broken in the short term, giving laser chip manufacturers high visibility for volume growth and stable or even increasing prices.

The "Silicon Photonics + CW Laser" solution offers significant advantages, providing domestic Chinese players a window to catch up and gain market share. An IIM institution report estimates that the global high-speed EML chip market is long-dominated by American and Japanese companies, with the top five suppliers holding a 70% market concentration. Leading players leverage deep technological积累 and strong customer ecosystem ties, creating a market characterized by "strong certification requirements, long cycles, and low substitutability." However, with the accelerating adoption of silicon photonics, the market格局 for CW lasers is not yet solidified, offering a rare window of opportunity for domestic manufacturers. On one hand, CW lasers, having decoupled the modulation function, only need to provide a stable light source. Their manufacturing process complexity is significantly lower compared to EML, requiring no modulator regrowth and having relatively relaxed parameter screening standards. This allows domestic manufacturers to potentially catch up in yield and achieve production capacity breakthroughs in a relatively short time. On the other hand, constrained by their own capacity bottlenecks, overseas leaders are prioritizing resource allocation to higher-margin EML chips, resulting in limited investment in the CW laser segment and a lack of deep binding with end customers. This creates conditions for domestic players to enter the North American supply chain. Based on company websites and investor communications, domestic manufacturers have successively achieved mass production and shipment of 70mW and 100mW CW lasers. Some players have even introduced 300mW high-power products targeting the CPO domain and are actively pursuing certification from North American cloud providers. They are well-positioned to rapidly capture market share during this critical period before the格局 solidifies.

Risk factors include slower-than-expected AI development; cloud provider capital expenditure falling short of expectations; delays in new product developments like single-channel 400G; technology path risks; and geopolitical risks.