Chey Warns of Deeper Chip Shortage in 2026 as AI Demand Surges

*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an "AI-based personalized news recommendation and summarization service" developed with support from the Korea Press Foundation. It selects and provides six customized news items by reader type.

1. Chey: 'Countries' Memory Demands Are Utter Chaos… Rapid Expansion Is Korean Chips' Lifeline'

- Key Summary: Chey Tae-won, chairman of the Korea Chamber of Commerce and Industry and chairman of SK Group, warned that AI-sector chip demand next year will surge at least 60-100% compared to this year, while there will be almost no increase in supply, widening the supply-demand gap further. Chey described the reality of governments exerting inter-state pressure to secure chip volumes as "utter chaos," and showed no signs of wavering despite U.S. Commerce Secretary Howard Lutnick's pressure for U.S. investment. Meanwhile, in response to the surge in AI-driven memory demand, SK Group will invest 400 trillion won to create a new semiconductor hub in the southwestern region and has decided to move up the completion of its Yongin cluster, originally scheduled for 2045, by 12 years. Chey also defined the current expensive semiconductor prices as "abnormal," stressing that Korea's semiconductor industry can be sustained and developed only if it increases supply to lower prices and grow the market, even at reduced margins.

2. CXL Surpasses HBM in Both Cost-Efficiency and Inference Performance… 'Market to Grow 7-Fold in Two Years'

- Key Summary: HBM and DDR, which are physically mounted directly onto GPUs (Graphics Processing Units) and CPUs (Central Processing Units), face structural limits that make further expansion difficult, and Samsung Electronics and SK hynix are developing CXL memory, which forms a separate memory tier, as an alternative. SK hynix improved AI inference efficiency by 35.7% with its IMTE architecture, structured as "HBM→DDR→CXL hybrid memory→SSD (Solid State Drive)," and Samsung Electronics announced that its CXL 2.0 memory can achieve 92% of the latest DDR5 performance. Accordingly, market research firm Yole Group projected that the CXL market will grow more than 652%, from $2.1 billion (about 3 trillion won) this year to $15.8 billion (about 23.5 trillion won) in 2028. Major big-tech firms including Nvidia, Intel, and AMD are also supporting CXL compatibility in their new chips launching in the second half, responding to the shift in the memory paradigm.

3. Bank of Korea Warns of 'Semiconductor Dutch Disease'… Counterarguments Cite 'Different Conditions Such as Exchange Rates'

- Key Summary: According to a report by the Bank of Korea's Economic Research Institute, first-quarter GDI (Gross Domestic Income) rose 13.2% year-on-year, marking a gap of 9.4 percentage points with the GDP (Gross Domestic Product) growth rate (3.8%) — the largest since statistics began in 1960. Unlike past "cost-reduction" booms driven by falling oil prices, this boom has a structural characteristic led by rising export prices, and major forecasting institutions observed that the capital expenditure of Samsung Electronics and SK hynix will surge from 75 trillion won last year to 120 trillion won this year and 150 trillion won next year. The BOK particularly warned of the possibility of "Dutch disease," in which semiconductor-concentrated growth erodes the foundations of other industries, citing that while the production share of IT manufacturing soared to 51%, the employment ratio in that sector stood at only 2.6%. On the other hand, there is also considerable counterargument that Korea's semiconductors are the result of decades of accumulated technological capital and that the exchange rate structure is different.

4. Bundling 8,000 Chinese Chips to Build a Supercomputer… Breaking Through U.S. Regulations With a Gum Strategy

- Key Summary: At the 9th World Artificial Intelligence Conference (WAIC) held in Shanghai on the 17th, Huawei unveiled the "Atlas 950 SuperPod," an AI computing system using a "SuperNode" method that connects 8,192 of its own "Ascend 950" chips with an ultra-high-speed communication network. Huawei claimed that this product leads the world in key metrics, including 6.7 times the total computing power compared to Nvidia's next-generation product "NVL144," and it is assessed that the "chip-bundling" strategy — connecting numerous chips as a substitute after U.S. export controls made securing cutting-edge chips impossible — actually worked. In addition, one day before the WAIC opening, China officially launched an international AI cooperation body involving 29 countries, including Russia and Brazil, ramping up "AI diplomacy" to counter the U.S.-led AI supply chain community "Pax Silica." Chinese President Xi Jinping announced a plan to provide 5,000 AI training opportunities to developing countries over the next five years.

5. Fearing Being Overtaken by China… Trump, Who Controlled AI Haphazardly, Speeds Up Self-Regulation

- Key Summary: The administration of U.S. President Donald Trump is accelerating the establishment of an independent regulatory body to review the safety of AI models, and it is reported that U.S. Treasury Secretary Scott Bessent is reviewing a plan modeled on FINRA (Financial Industry Regulatory Authority, a private self-regulatory body overseeing securities firms). This is interpreted as a measure following growing dissatisfaction in the AI industry over the application of regulations without clear standards, such as Anthropic's temporary suspension of model services. Meanwhile, according to AI model marketplace OpenRouter, as of the 13th of this month, Chinese AI models — Tencent, Xiaomi, DeepSeek, MiniMax, and Zhipu AI — occupied all of the top five spots in weekly token usage, and Chinese AI startup Moonshot's open-weight model "Kimi K3" put out its own assessment that, with 2.8 trillion parameters, it outperformed the flagship models of Anthropic and OpenAI in benchmarks. There is an observation that if the criteria for choosing AI models shift toward price-to-performance, the overall investment strategy concentrated on developing hyperscale AI models and computing infrastructure could be reexamined.

6. Urgent to Digitize 'Disappearing Tacit Knowledge'… Only 0.1% of SMEs Have Adopted Manufacturing AI

- Key Summary: According to the National Ppuri Industry Promotion Center's "2025 Ppuri Industry Survey," among workers in the root industries (744,543 people) as of 2024, skilled workers aged 50 and over accounted for 36.8% of the total, while those under 30 accounted for only 10.3%, indicating that the disruption in technology transfer is deepening. In the Ministry of SMEs and Startups' "2024 Smart Manufacturing Innovation Survey," those that adopted manufacturing AI