Legend Holdings (03396) has a multidimensional layout in optical interconnection technology.

When the growth rate of GPU performance shifts from "Moore's Law" to "new Moore's Law," chip-to-chip communication efficiency has become a core variable determining the efficiency of intelligent computing clusters. On May 6th, NVIDIA reached a long-term commercial and technical cooperation agreement with Corning. According to the agreement, Corning will build three advanced manufacturing facilities in North Carolina and Texas specifically for producing optical interconnect solutions for NVIDIA. NVIDIA, on the other hand, invested $500 million to acquire two immediately exercisable options to purchase Corning common stock. This is the latest investment move by NVIDIA in the field of optical interconnects this year. In March 2026, NVIDIA invested a total of $6 billion in 30 days, investing $2 billion each in Lumentum, Coherent, and Marvell. The above intensive investments all point to the field of optical interconnects and AI computing power infrastructure - a technology that uses optical waves as carriers and optical fibers as channels, with the "ultra-high bandwidth, ultra-low power consumption" core advantage becoming the key to breaking through the AI computing power bottleneck. Bottleneck emerging: The ceiling of electric interconnects Bandwidth walls, power walls - are hindering the further upgrading of AI infrastructure. In the past two years, global AI computing power has increased by about 60,000 times, but the "blood vessels" supporting them are severely lagging behind: traditional copper cable electrical interconnect bandwidth has only increased by 30 times, creating an ever-increasing gap. At the same time, as GPU power consumption continues to rise, electrical interconnect power consumption may further increase. For the industry, this will not only increase data center operating costs but also bring heat dissipation pressures. Even more serious, long-distance transmission losses, frequent electromagnetic interference, and reaching the integration density limit, these real problems pose unprecedented transmission challenges for training models with billions of parameters and parallel computing with tens of thousands of card clusters. Technical breakthrough: Optical interconnects enter an accelerated development period While electrical signals struggle to traverse copper wires, photons are already shuttling back and forth in optical fibers close to the speed of light. With its physical nature, optical interconnect technology exhibits outstanding advantages such as ultra-high bandwidth density, extremely low power consumption, ultra-low transmission latency, strong interference immunity, and so on. Leading technology companies have keenly captured the technological advantages of optical interconnects and are actively making commercial arrangements. - AMD announced the development of MRM-based CPO (co-packaged optics) solutions for the next generation Instinct MI500 AI accelerator, establishing an industry chain division of "AMD design + GeX manufacturing + J/Moonlight packaging", expected to be launched in 2027. - NVIDIA's actions are even faster, with a total investment of $6 billion in 30 days, all directed towards optical interconnects and AI computing power infrastructure. Its Rubin Ultra architecture will prioritize the use of CPO solutions, while the longer-term Feynman generation plan will fully transition to optical interconnect technology. - TSMC's silicon photon engine COUPE platform, based on the SoIC-X chip stacking technology, directly stacks the electrical chip with the photonic chip in 3D, completed small-scale pluggable optical module verification in 2025 and began large-scale production in 2026; at the same time, the platform will also integrate COUPE optical engines into CoWoS packaging in the form of CPO, a key step in the implementation of CPO. - Samsung Electronics and LG Innotek of South Korea have formally transitioned from the concept research stage of CPO technology to early development, expected to integrate various components on semiconductor substrates to enable the final packaged semiconductor product to have CPO functionality. Domestic practice: Multi-dimensional layout of "production, study, research, and investment" The accelerated landing of technology has created a huge market space. According to a Goldman Sachs research report, the global addressable market for optical interconnects will increase from the current $15 billion to $154 billion in 2027-2028, with an overall increase of 9 times. Faced with the industry development opportunities, the LEGENDHOLDING system, with its long-term accumulation in cutting-edge technology fields, has made forward-looking layouts in the field of optical interconnects. - The "Peking University-LEGENDHOLDING Advanced Photonics Integration Technology Joint Laboratory" jointly established by LEGENDHOLDING and Peking University is conducting joint research in areas such as photon integration process exploration, low-power optical engine prototype verification, and optical interconnect application research for intelligent computing centers. In March 2026, the first academic paper of the joint laboratory was published in the internationally renowned academic journal "Photonics," where the research team overcame the technical limitations of short-wavelength optical grating couplers in terms of coupling efficiency and process tolerance, and related technologies have been patent-protected. - Junlian Capital, Lenovo Star, and Lenovo Ventures jointly invested in Turing Quantum as a pioneer in the field of photonic quantum computing, possessing full-stack technology capabilities from chips and algorithms to complete system systems. Recently, Turing Quantum and the leading AI network interconnect enterprise Qimeng Moore reached a deep strategic cooperation, jointly developing and promoting the next-generation optical interconnect OIO technology project, building a light-electric fusion hardware system covering classical computing to quantum computing. This collaboration is bringing key technical capabilities of light quantum computing compatibility downwards, enabling the optical interconnect field, pushing computing past bottlenecks, and building a new paradigm of light-electric fusion computing. - Guanglianchip Science and Technology, an enterprise invested by Junlian Capital, is a domestic innovative leader in the field of OIO optical interconnects, focusing on intra-chip optical interconnects for high-performance computing. It has accelerated the transition of OIO technology from research and development to practical deployment, its multi-channel OIO silicon optical chip and optical engine products can achieve direct light emission from computing chips, significantly increase chip-to-chip transmission bandwidth while reducing transmission energy consumption, achieving a thousand-fold increase in bandwidth efficiency compared to data center optical modules, greatly enhancing the training and inference efficiency of AI clusters, helping to land AI large model and promoting the intelligent computing center towards all-optical interconnection. - Xili Optical Electronics, invested by Lenovo Ventures, has a deep international research background and industrialization capabilities, covering key areas such as material understanding, wafer-level processes, chip design, device engineering, and industrial cooperation. The company's founder/chief scientist, Professor Wang Cheng, is one of the main pioneers of global thin-film lithium niobate photonic technology and the core inventor of thin-film lithium niobate electro-optic modulators. Relying on the original innovation accumulation and the engineering transformation capabilities of the industrialization team, Xili Optical Electronics is accelerating the application of thin-film lithium niobate technology from the laboratory to AI optical interconnection and next-generation communication networks. The technological prospects of optical interconnects are now highly clear. Its value lies not only in satiating the thirst for computing power but also in defining the future computing architecture. When the bottleneck of data transmission is overcome, the fusion of distributed computing, in-memory computing, and even quantum computing can truly unleash their potential, injecting strong momentum into AI infrastructure.