Jibang Consulting: AI Data Centers Explode Optical Communication Laser Shortage Nvidia (NVDA.US) Strategic Layout Reshapes Laser Supply Chain Pattern
The Zhitong Finance App learned that according to the latest research from TrendForce Jibang Consulting, with the development of data centers towards large-scale clusters, high-speed interconnection technology has become the key to determining the maximum efficiency and large-scale development of AI data centers. In 2025, the world will have 24 million optical transceiver modules above 800G, and it is estimated that in 2026 it will reach nearly 63 million sets, a growth rate of 2.6 times.
TrendForce Jibang Consulting pointed out that due to the huge demand for high-speed optical transceiver modules above 800G, it has caused serious supply bottlenecks in the upstream laser light source in the supply chain. In particular, Nvidia (NVDA.US) monopolizes the production capacity of EML laser chip suppliers due to strategic considerations. As a result, EML laser delivery is already scheduled after 2027, causing a shortage of supply in the laser light source market. Various optical transceiver module manufacturers and CSP customers of terminals are also seeking more suppliers and solutions due to the shortage of laser light sources, which is driving changes in the laser industry pattern.
Nvidia strategically monopolizes EML laser chips
In addition to being used to transmit VCSEL lasers over short distances, lasers in medium- and long-range optical transceiver modules can currently be divided into two types: EML lasers and CW lasers. Among them, because EML lasers integrate signal modulation functions within a single chip, the production threshold is extremely high and the optical components are complex, so there are only a few global suppliers. Currently, the main suppliers include Lumentum, Coherent (Finisar Finisar), Mitsubishi (Mitsubishi), Sumitomo (Sumitomo), Broadcom (Broadcom), etc.
Due to the emergence of very large data centers, EML lasers with a longer penetration distance and stable signal have become a key strategic material. Coupled with the slow progress of mass production of Nvidia's silicon optical/CPO, large-scale reliance on pluggable optical transceiver modules is still required in the short term to meet the needs of GPU clusters. Therefore, in order to ensure safe supply, production was contracted to its EML laser chip supplier, resulting in a tight supply of EML laser chips on the market.
CW Laser, the new darling of big cloud manufacturers and production capacity competition
In contrast, CW (continuous wave) lasers are only responsible for providing a constant light source and using silicon photonics (Silicon Photonics) chips manufactured in semiconductor foundries as external modulators to convert electrical signals into optical signals for transmission. Therefore, there is no need to integrate modulation functions on laser chips, and the chip structure is relatively simple. This is also the reason why CW laser solutions using silicon photonic technology have become the preferred alternative for major CSP manufacturers when EML lasers are in short supply.
Although there are relatively many manufacturers investing in CW lasers, supply is not as tight as EML lasers. However, in the face of huge demand for high-speed AI transmission, the expansion of CW laser production capacity is also limited by the delivery of production equipment and cannot be rapidly expanded, so it is difficult to meet huge customer demand in the short term. In order to meet customer reliability requirements, the late-stage chip cutting and testing process also consumes quite a lot of manpower and related resources. Therefore, in addition to making the most critical laser chip cutting and aging test process, many laser manufacturers now selectively outsource the late-stage laser chip cutting and aging test process to other relevant laser manufacturers, causing the production capacity of related laser supply chains to gradually tighten, and they are also launching production expansion plans one after another.
Demand for high-speed PD surged at the same time, and the Leijing factory ushered in spillover business opportunities
As for optical transceiver modules, in addition to the laser light source at the emitter end, a photodiode (PhotoDiode, PD) is also required as an optical receiving element. In order to match the transmission speed of a higher laser light source, various PD manufacturers are also currently investing in the development of high-speed PDs that can receive 200G transmission signals. Manufacturers such as Coherent, Macom, Broadcom, and Lumentum have also launched 200G high-speed PDs.
Since high-speed PD and EML are the same as CW lasers, they all use INP (indium phosphide) substrates and then decrystallization. In a situation where there is a shortage of laser light sources, laser manufacturers tend to allocate most of their production capacity to produce laser light sources, and at the same time hand over INP Leijing's production to professional Leijing foundries such as IET-Interlei and Xinxin through outsourcing.
TrendForce Jibang Consulting believes that, driven by huge demand for AI, in addition to a serious shortage of memory, high-speed transmission is also driving the supply tightening in the upstream laser industry. In particular, although Nvidia's EML laser monopoly strategy guarantees its own shipping safety, it also unexpectedly accelerated the adoption of CW lasers and silicon photonics technology by non-Nvidia camps. At the same time, this battle for production capacity is reshaping the division of labor in the supply chain, bringing significant growth momentum to related supply chains with advanced compound semiconductor epitaxial and process capabilities.