Stocks of companies related to artificial intelligence (AI) semiconductors have sharply declined due to concerns over low-cost AI shocks from China and fears of a peak-out in the market. However, analysts suggest it is premature to declare the end of the memory supercycle, as the AI industry is expanding beyond generative AI and agents into physical AI, AI factories, and data centers, leading to explosive demand growth.
On July 20, both the KOSPI and KOSDAQ markets triggered sell-side circuit breakers. Samsung Electronics and SK Hynix both experienced declines, contributing to a broader drop in indices. This downturn reflects a combination of global semiconductor stock adjustments and concerns over the Chinese AI shock that accumulated over the three-day holiday weekend.
The stock market trends diverge somewhat from corporate performance. Samsung Electronics reported record earnings for the second quarter, while SK Hynix is also expected to post strong results based on demand for high-bandwidth memory (HBM). The peak theory circulating in the market has two main arguments. The first is the potential slowdown in AI investments by U.S. big tech companies, raising doubts that rising costs for data centers and other investments may outpace revenue generation. If big tech slows its capital expenditures next year, demand for GPUs and HBM could also weaken.
The second concern is price pressure from China. CXMT's expansion in DRAM production is seen as a variable that could disrupt general memory prices. Low-cost AI models like the 'Kimi K3' from China's Moonshot AI could also dampen demand for expensive GPUs and HBM. However, this situation is viewed more as a reassessment of AI investment profitability and memory pricing rather than the end of the supercycle.
In fact, industry indicators have not yet tilted toward a peak-out. The Bank of Korea recently assessed the current global semiconductor market expansion as the strongest since the 2000s. A key point noted by the Bank of Korea is the evolution of AI. AI is expanding from text-based large language models to multimodal models that process both video and audio, which will drive demand for high-performance memory and computing semiconductors. The spread of physical AI is also expected to increase the demand for high-performance computing and low-power semiconductors.
Market research firm Gartner has also projected additional growth in the semiconductor market. Gartner forecasts that global semiconductor revenue will rise from $805.3 billion last year to $1.3202 trillion this year, and further to $1.5545 trillion next year. This year's growth rate is expected to be 64%, the highest in 20 years. Notably, memory revenue is projected to increase nearly threefold from $216.3 billion last year to $633.3 billion this year, with expectations of reaching $748.1 billion next year.
The Boston Consulting Group (BCG) views data centers as the foundational infrastructure for the spread of physical AI. BCG analyzes that AI infrastructure will become a strategic asset supporting seamless AI services in manufacturing and industrial settings. As AI technology transitions from screen-based services to real-world applications, demand for data centers and semiconductors will inevitably grow.
Jung So-young, CEO of NVIDIA Korea, stated, "Energy, semiconductors, and data centers are currently in a serious bottleneck situation. Supply is unable to keep up with demand, leading to limitless advancements in AI."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.