[Edaily Reporter SOYEON KIM ] The competition in the artificial intelligence (AI) semiconductor sector is expanding from simply enhancing the performance of high-bandwidth memory (HBM) to maximizing the efficiency of the entire AI system. Not only the performance of individual semiconductors but also how quickly and efficiently data is exchanged between semiconductors and servers has emerged as a new competitive factor. SamsungElectronics and SK hynix are also responding to the next-generation AI infrastructure market by expanding their research and development into optical interconnect technology, which transmits data using light.

According to industry sources on the 9th, U.S. semiconductor company Marvell recently agreed to acquire optical components manufacturer Lumentum for approximately $115 billion (based on enterprise value). This strategy aims to secure a leading position in the AI data center market by combining data interconnect semiconductors with laser and optical communication technologies.

As AI data centers have evolved into massive systems connecting thousands of graphics processing units (GPUs), data transmission has emerged as a new bottleneck. Optical interconnect technology transmits data using light instead of copper-based electrical signals, thereby increasing speed and reducing power consumption. Because it can quickly transmit large volumes of data while minimizing power consumption and signal loss, it is gaining attention as a core technology for next-generation AI data centers.

SamsungElectronics officially announced its entry into the silicon photonics foundry business last March and is moving forward with the establishment of a photonic integrated circuit (PIC) production platform. Silicon photonics is a technology that utilizes existing semiconductor manufacturing techniques to integrate optical devices—which transmit light or convert electrical signals into light—onto a single chip. Through this initiative, SamsungElectronics aims to unveil an optical engine in 2027 and provide a turnkey Common Photonic Packaging (CPO) service by 2029.

An optical engine is a device that combines a photonic integrated circuit (PIC), which transmits data using light, with an electronic integrated circuit (EIC), which processes electrical signals. CPO is a technology that integrates these optical engines with core semiconductors—such as GPUs or network switches—into a single package. By reducing the distance electrical signals must travel, it can lower the power consumption required for data transmission.

SamsungElectronics has already secured customers for its optical communication semiconductors and is currently discussing CPO commercialization projects with major clients. The company’s strategy is to combine its business capabilities—ranging from memory to foundry and advanced packaging—with optical interconnect technology to enhance the data transmission efficiency of AI systems as a whole.

Last August, SK hynix published a CPO technology roadmap—developed in collaboration with researchers from a U.S. university—in the international academic journal *Nature Electronics*. The roadmap proposed an architecture in which multiple AI accelerators share large-scale memory resources using optical interconnects. This demonstrates that the scope of technology development for memory companies is expanding beyond improving the performance of individual semiconductors to optimizing the data transmission architecture of the entire AI system.

An industry official stated, “The focus of competition is shifting toward improving the efficiency of the entire data center,” adding, “Resolving data bottlenecks and supporting the efficient operation of AI systems will become the new measure of technological competitiveness.”