Concerns about a potential downturn in the semiconductor supercycle have emerged amid discussions among global tech giants about regulating the pace of AI development. However, the domestic semiconductor industry appears to be unaffected, buoyed by robust demand for general memory driven by a surge in data processing for AI services.
According to industry reports, global DRAM sales in the second quarter of this year soared by 59.5% from the previous quarter, reaching $15.473 billion (approximately 208 trillion won). Samsung Electronics maintained the top position with a 39.4% market share, bolstered by the mass production of sixth-generation high-bandwidth memory (HBM4). HBM exports also surged by 293% compared to the same period last year, with the average export price rising by 31% in just two months to $76, highlighting the strong performance of high-value memory products.
Market research firm TrendForce noted, "The rapid increase in HBM demand for AI servers, coupled with rising prices for general DRAM, has quickly boosted overall DRAM industry revenues."
Experts emphasize that the recent calls for regulating AI development do not indicate a decline in the semiconductor market. They argue that the financial adjustments made by major AI companies, due to infrastructure costs and pressures to improve profitability ahead of public offerings, should not be interpreted as a decrease in demand.
Kim Yang-pyung, a researcher at the Korea Institute for Industrial Economics and Trade, stated, "The discussions around AI speed regulation are largely internal adjustments aimed at managing astronomical costs accumulated during the training of large models and restructuring profit frameworks. Interpreting this as a signal of a demand cliff or peak-out for the entire semiconductor industry is an excessive concern disconnected from on-the-ground data."
Despite warnings about AI investments, exports of domestic semiconductor chips have actually increased. According to the Ministry of Trade, Industry and Energy, South Korea's semiconductor exports reached a record $46.7 billion last month, marking a 68.7% increase compared to the same period last year. Instead of a contraction in memory orders, which had been feared as a turning point for AI investments, the trend has been bolstered by rising prices for key products.
As the focus of the AI market shifts from training large models to operating practical services, the portfolio of K-Semiconductors is diversifying. According to global market research firm Deloitte, the share of inference in total AI operations used for actual services has increased to 66% this year, surpassing the scale of training operations by nearly double. As AI moves beyond the training phase to practical user applications, demand for high-performance HBM, as well as general DRAM products like GDDR7 and LPDDR5X, is surging.
Lee Jong-hwan, a professor in the Department of Semiconductor System Engineering at Sangmyung University, commented, "While the frenzy and speed competition surrounding large model construction may temporarily slow down, the real-time inference traffic generated by millions of users using AI in their daily lives has entered a significant growth phase. The demand for memory, which was previously concentrated on training chips, is expanding across the entire market, including general DRAM, further solidifying the performance of South Korean semiconductors."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.