In response to concerns about a peak in the semiconductor market, the artificial intelligence (AI) industry argues that excess demand for AI and semiconductors will continue at least until 2030. However, China's low-cost strategy, backed by national support, is seen as a potential variable that could impact the market landscape.
On July 20, Jensen Huang, CEO of NVIDIA, announced plans for AI infrastructure development in Japan, stating, "The global AI infrastructure buildout is just beginning," and asserted, "This cycle will last for more than the next decade." His comments dismiss concerns about an early slowdown in the anticipated supercycle.
In South Korea, there are calls to accelerate infrastructure development to meet the excess demand for semiconductors. Ryu Jin, chairman of the Korea Economic Association, noted that "70-80% of domestic AI queries are processed by overseas data centers," emphasizing the urgent need to expand domestic infrastructure centered around semiconductors.
The recent imbalance in memory supply stems from a paradigm shift due to the transition to AI. The Bank of Korea has also reported that "the current expansion phase is based on structural demand driven by AI proliferation," predicting that the duration of this boom will significantly exceed that of past cycles.
While aggressive investments from global tech giants continue, the supply of memory to support these efforts remains sluggish. Considering the timelines for new fabs from Samsung Electronics and SK Hynix to stabilize yields, it is expected that supply will only improve around 2030. If new fabs are not secured in time, supply shortages could worsen after 2032.
Hong Sang-jin, a professor at Myongji University, stated, "The memory production structure is being restructured around high-performance products like HBM, making it impossible to simply increase supply," adding that big tech companies are strengthening partnerships with both domestic firms through long-term supply contracts to secure volumes.
China poses a potential disruptor in this market trend. Changxin Memory Technologies (CXMT) achieved a global market share of 8% last year, ranking fourth in the industry. With an upcoming IPO on the Shanghai Stock Exchange on July 27, CXMT plans to invest the funds into expanding its monthly wafer production capacity to 500,000 by 2028. Yangtze Memory Technologies (YMTC), which focuses on NAND flash, has also increased its market share to 13% in the first quarter and is accelerating its market penetration by advancing the mass production timeline of its third factory in Wuhan to the end of this year.
Not only in memory, but China is also pursuing low-cost strategies in AI models. Moonshot AI's "Kimi K3" has demonstrated performance comparable to top U.S. models at a lower cost. This strategy aims to dominate the global AI ecosystem by combining hardware manufacturing capabilities with software cost-effectiveness.
While China's memory technology still lags behind South Korea, it aims to enter the market by targeting the production of next-generation advanced products like HBM.
Lee Jong-hwan, a professor at Sangmyung University, warned, "China is likely to flood the general DRAM market with low-cost products, driving down overall prices," and emphasized that South Korean companies must maintain market leadership through overwhelming technological advantages rather than price competition.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.