CITIC SEC: AI drives the rise in both quantity and price of laser chip, focusing on new opportunities for domestic production.

CITIC SEC released a research report stating that the next stage of AI computing clusters is moving from "stacking computing power" to "competing in network efficiency." Laser chip is the core device of optical module, determining the electro-optical conversion efficiency and product generation of the optical module. The laser chip industry will benefit from the value increase brought by the expansion of Scale-out/Scale-up network super nodes and technological generation transition, ushering in huge development opportunities. At the same time, due to strong market demand and short-term rigid supply, the laser chip market is in a phase of rapid growth and supply-demand imbalance. The advantages of leading manufacturers are continuously highlighted. Bullish on the future development potential of leading laser chip companies. CITIC SEC's main points are as follows: Laser chip is the core value component of the optical module, with CW and EML as the dominant solutions at high speed rates. The laser chip is an optoelectronic source chip based on III-V group semiconductor materials, with the core function of converting high-speed electrical signals into light signals of a specific wavelength. In the AI data communication market, laser chips are mainly divided into three categories: VCSEL, EML, and silicon photonics, with the core DRIVE always focusing on "single-channel rate increase" and "transmission distance extension." In the era of 100G and earlier, data centers mainly used short-range interconnects within hundreds of meters. VCSEL dominated due to its low power consumption and cost. As we enter the era of 400G/800G, the single-channel rate jumps to over 100G, with VCSEL limited by the difficulty in breaking the 30GHz bandwidth due to multi-mode effects, and the interconnect distance of AI clusters extends to 500m-2km, highlighting the constraints of multi-mode fiber dispersion, causing VCSEL solutions to exit the mainstream. EML, with DFB+EAM (electro-absorption modulator) single-chip integration, quickly became the dominant solution for 800G/1.6T. In the era of 1.6T, the silicon photonics solution integrates modulators, detectors, etc., into a single PIC chip with mature CMOS technology, significantly reducing packaging complexity and power consumption. The CW lasers can achieve 1:2 or 1:4 reuse through spectral splitting, directly diluting the overall cost and becoming the market mainstream. The era of 1.6T is a key turning point for the silicon photonics solution to become the "core main force" from being an "optional choice," and the strategic value of CW lasers under the silicon photonics solution is upgrading from "supporting components" to "core bottleneck components." The high-speed laser chip market driven by AI is experiencing phenomenal growth, with a strong trend of price and volume increase. From the demand side, the demand for AI large model training and inference is continuously increasing, with significant capex upgrades from global cloud vendors, expanding AI computing clusters, and interconnect efficiency becoming the core bottleneck of cluster computing. As the cluster size expands from thousands of cards to tens of thousands or even millions of cards, the ratio of optical modules per card has increased significantly from 1:2.5 to 1:5 or even 1:12, and the demand for optical modules is expanding much faster than the growth of GPUs, leading to a significant increase in the demand for laser chips. Meanwhile, optical interconnects are accelerating penetration into traditional copper connections, further opening up incremental space. From the supply side, the ramp-up cycle of laser chip production lasts 18-24 months, with the core MOCVD equipment currently monopolized by two US and German vendors, leading to significant short-term production rigidity. The production capacity of leading laser chip manufacturers is locked in until 2027, with some customer orders even visible up to 2028. According to CITIC SEC's estimation, the market size of high-speed laser chips (100G and above) is expected to jump from $930 million in 2025 to about $4.06 billion in 2026, a year-on-year growth of over 338%, further expanding to about $7.10 billion in 2027, a cumulative growth of over seven times in two years. Among them, the supply gap for 200G EML is as high as 20%, and the gap for 400mW CW lasers is 17%. The seller's market structure is difficult to break in the short term, and laser chip manufacturers have a high visibility of expected volume increases and price stability or even price hikes. The "silicon photonics + CW laser" solution has significant advantages, and domestic rapid catching up in CW laser market share is expected. According to IIM Institute's "Global and China CW DFB/EML Laser Chip Market Industry Chain Analysis Report" (December 2025), the global high-speed EML chip market has long been dominated by US and Japanese companies, with the top five suppliers accounting for 70% of the market share. Lumentum, Broadcom, Sumitomo Electric, and other leading manufacturers have formed a market competition pattern of "strong certification, long cycle, weak replacement" based on their deep technological accumulation and customer ecosystem. However, with the accelerating penetration of silicon photonics solutions, the market structure of CW lasers has not solidified yet, providing a rare window of opportunity for domestic manufacturers. On one hand, after decoupling the modulation function, CW lasers only need to provide a stable light source, significantly reducing manufacturing complexity compared to EML, without the need for regrowth of the modulation area, with relatively loose parameter screening standards, domestic manufacturers are expected to achieve yield catch-up and capacity breakthrough in a short period of time; on the other hand, overseas leading manufacturers, limited by their own production capacity bottlenecks, prioritize resources on higher-margin EML chips, with limited investment in the CW laser field and no deep customer binding, creating conditions for domestic players to enter the North American supply chain. According to various manufacturers' websites and investor logs, domestic manufacturers have successively achieved mass production and shipment of 70mW and 100mW CW lasers, some manufacturers have even launched high-power products of 300mW for the CPO field, and actively promoting certification with North American cloud vendors, aiming to quickly seize market share in the critical stage when the market structure is not solidified. Risk factors AI development does not meet expectations; cloud vendors' capital expenditures are lower than expected; progress of new products such as single-channel 400G is slower than expected; technology path risks; geopolitical risks.