SK Hynix is leveraging optical interconnect technologies to maintain its leading position in the HBM market against intensifying competition.

SK Hynix appears to be positioning optical interconnects and co-packaged optics (CPO) as the critical technology to breach the AI memory wall. A research paper authored by the South Korean memory leader alongside academics from the University of Virginia, Nanyang Technological University, and the Massachusetts Institute of Technology (MIT) examines how optical compute interconnects can support next-generation AI systems. The researchers advocate replacing copper interconnects with optical links, describing it as “the most promising path toward future scalability.”

While CPO technologies are increasingly adopted to accelerate chip-to-chip links within compute trays and between switches, SK Hynix and its research partners envision extending these interconnects all the way to the memory interface. The paper outlines an “optics-centric architecture” that utilizes photonic interposers—chip-packaging substrates that route data through both electrical signals and light—to directly link memory pools and processors. This approach aims to maximize data movement efficiency across computing systems. By enabling multiple AI accelerators to share expansive memory pools at speeds significantly exceeding conventional storage links, the architecture would facilitate highly efficient, scalable system-level data movement.

“Extending optical interconnects to the memory interface would overcome the physical constraints surrounding compute chips, removing limitations on both memory capacity and the number of electrical connections,” University of Virginia Professor Kyusang Lee explained. “It would also allow multiple AI accelerators to share a large memory pool, which would make AI infrastructure far more adaptable as AI models continue to grow in scale.”

The ongoing memory shortage, which has driven hardware prices to record highs, is largely fueled by AI data center developers. Although AI accelerators have seen exponential performance gains in recent years, high-bandwidth memory (HBM) struggles to keep pace. HBM consumes substantially more silicon wafer capacity per gigabit than conventional memory, creating significant performance bottlenecks. Consequently, AI infrastructure operators have been aggressively purchasing available supply to satisfy their high-performance computing stacks.

Efforts to break through this memory wall have spurred development of faster, lower-power alternatives such as 2T0C from South Korea’s Electronics and Telecommunications Research Institute, as well as chiplet-based interconnects designed to deliver low-latency connectivity between compute engines and memory. SK Hynix is also supporting a potential HBM alternative in the form of Sandisk-developed high-bandwidth flash. However, given its dominant position in the HBM market, SK Hynix’s investment in CPO serves as a strategic defense mechanism to maintain the relevance of its flagship product. The intense demand for HBM previously propelled the South Korean memory giant’s market valuation to eclipse $1 trillion and fuel the largest ever market launch by a foreign firm on the Nasdaq. In recent trading activity, the South Korean chip giant’s shares jumped almost 30% on the Nasdaq and 13% on the KRX.