The demand for memory semiconductors is rapidly expanding beyond AI data centers to encompass supercomputers, AI PCs, and robotics, driven by the proliferation of artificial intelligence (AI). The applications for advanced K-memory, which has primarily focused on AI accelerators like graphics processing units (GPUs), are expected to broaden significantly.
On June 8, Jensen Huang, CEO of NVIDIA, emphasized the importance of Korean memory semiconductors during his third day in South Korea. All four new products recently unveiled by NVIDIA require substantial memory input. The next-generation AI accelerator, 'Vera Rubin,' will utilize a vast amount of high-bandwidth memory (HBM), while the new central processing unit (CPU) 'Vera' and the AI PC platform 'RTX Spark' will be equipped with a significant amount of low-power, high-performance DRAM (LPDDR5). The robotics platform 'Jetson Thor' is also in the process of developing customized memory.
During a dinner meeting on June 5 in Hongdae, Huang stated, "Next year, four new products will be launched simultaneously, and we will need a lot of Korean memory semiconductors," adding, "We will strengthen our partnerships with Korean companies like SK Hynix and Samsung Electronics."
A particularly noteworthy area is physical AI. As robots and autonomous manufacturing systems become more widespread, the demand for memory to support real-time data processing and AI inference will inevitably increase. According to a report from the Korea Automotive Technology Institute, the current DRAM capacity used in standard vehicles is around 16GB, but level 4 technology could require up to 300GB, with similar capacities expected for robots.
AI factories are also emerging as significant demand sources. As AI systems increasingly control overall factory operations, the demand for memory in data centers, robots, and edge devices is expected to surge. The need for HBM, as well as DRAM for CPUs, LPDDR, and memory for robots, positions these technologies as key infrastructure driving next-generation memory demand.
The collaboration between SK Hynix and NVIDIA to co-develop next-generation memory is seen as a proactive strategy to respond to this market expansion. Their partnership, which initially focused on HBM, is now extending into the realm of customized memory semiconductors for physical AI and AI factories. Huang remarked, "SK is our largest memory partner," and noted that they are jointly designing a roadmap to advance NVIDIA's architecture and SK's memory technology.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.