ReadingResearchRadarInvestment frameworkAI Company Primary Sources
Sign in / Sign up
← Back to reading
NewsStructural5 min readDisconfirmAI Editorial

Analyst Richard Gray: How the 'Memory Discount' on AI Storage Chips Amplifies Global Market Risk

In an analysis published in November 2026, analyst Richard Gray introduces the concept of a 'memory discount': forward price-to-earnings ratios for memory chip companies are systematically lower than those of other stocks on the same exchange, both in the United States and South Korea. He argues that this discount, combined with the extreme industry concentration in the South Korean stock market, structurally undervalues the most critical physical layer of AI in financial pricing. The catch-up efforts by China's CXMT and ongoing debates about storage cycles imply that current valuation logic, built on shortages, faces fundamental re-evaluation risks.

2026-11-01 21:33 GMT+8

Richard Gray is an analyst at a foreign policy think tank in Washington, D.C., United States. In an analysis published in November 2026, he points out that from March 2025 to June 2026, the KOSPI index of South Korea rose by 241%, adding nearly $3.3 trillion in market capitalization, exceeding the total cumulative increase accumulated over the index's 43-year history; it then evaporated nearly $1.1 trillion during a bear market from June 22 to July 16. He believes this is not just a South Korean issue, but a structural signal of the global AI financial architecture: Samsung Electronics and SK hynix of South Korea control 83% of the global high-bandwidth memory (HBM) market and 63% of the dynamic random-access memory (DRAM) market, making the South Korean stock market functionally equivalent to a highly leveraged single node within the AI supply chain.

Gray proposes two interwoven explanatory frameworks. The first is the 'Korea discount'—South Korean equities are structurally undervalued due to lagging financial development and weak governance, with MSCI still classifying South Korea as an emerging market on the grounds of limited offshore convertibility of the won, excessive requirements for foreign securities, and insufficient investment instruments. The second, which he considers more critical, is the 'memory discount': forward price-to-earnings ratios for memory chip companies are systematically lower than those of other stocks on the same exchange, whether in the United States or South Korea. In South Korea, SK hynix and Samsung are valued lower than Naver, Hanwha Ocean, and Kakao; in the United States, despite being in a capital market significantly deeper than South Korea's, Micron's price-to-earnings ratio is similar to that of SK hynix and Samsung.

Sign in / Sign up
ReadingResearchRadarInvestment frameworkAI Company Primary Sources

The root of this discount lies in the historical cyclicality of the storage industry—where competition has primarily revolved around price rather than quality, and demand fluctuations have compressed the number of players. Whether the AI boom has truly changed this nature remains a subject of debate. Samsung has locked 70% of its storage capacity into long-term agreements through 2030, SK hynix CEO Kwak Noh-Jung stated that the shortage could persist until 2030, and JPMorgan estimates that DRAM prices rose by more than 400% from 2024 to 2026. However, new storage capacity planned in South Korea is set to come online in 2030, coinciding exactly with the end of the shortage predicted by Kwak. Gray argues that if storage re-commoditizes, CXMT becomes a strong HBM competitor, and prices collapse, the balance sheets of these national champions will face fundamental pressure.

Among storage players, the only one that does not fit the 'memory discount' is China's ChangXin Memory Technologies (CXMT). After listing in July 2026, its price-to-earnings ratio exceeded that of Baidu, Alibaba, Tencent, and Xiaomi. Gray cites Zongyuan Zoe Liu, Senior Fellow at the Council on Foreign Relations in the United States, who noted that the Chinese government is encouraging long-term patient capital, such as insurance and pension funds, to enter the equity market to support national champion enterprises. Consequently, CXMT's market capitalization is both a financial position and a reflection of national industrial policy. CXMT currently produces fifth-generation HBM3E, two generations behind the seventh-generation HBM4E samples being sent by SK hynix and Samsung, but its share of the DRAM market has grown from 3% in Q1 2025 to 10% in Q2 2026, with profits reaching $22.4 billion in the first half of the year and sales increasing by 874%.

Gray juxtaposes three sets of dynamics: U.S. frontier laboratories have high valuations but low revenue; Chinese firms have suppressed valuations due to state support (as they do not need to rely on market pricing to access capital); and South Korean memory giants have exponentially growing revenues yet still carry a discount. CXMT is the exception in this landscape—as China's only enterprise close to the frontier in storage, state support is reflected in its high valuation. The simultaneous existence of these three dynamics means that the physical layer of AI has not yet been fully reflected in its irreplaceability in financial pricing. Samsung and SK hynix account for more than half of the KOSPI market capitalization in the South Korean stock market (compared to the Magnificent Seven accounting for 37.2% of the S&P 500 in the United States), and this concentration means that any industry-level volatility is directly transmitted as a national-level financial event. From June to July 2026, Evercore wrote that South Korea had become the 'tail that wags the dog' of global asset markets, with the correlation between the KOSPI and the Nasdaq 100 reaching 0.95. South Korean retail investors entered heavily via single-stock leveraged ETFs, with such trades accounting for 84% of total volume in June; the Financial Services Commission of South Korea subsequently suspended the sale of stock-linked leveraged ETFs.

Gray argues that the combination of the memory discount and the Korea discount creates a fragile system: upside depends on storage performance, while downside is cross-industry—the shipbuilding, chemicals, construction, and steel sectors all fell synchronously during the 2026 bear market. He suggests that South Korea should indefinitely ban single-company, dual-company, or single-industry ETF trading and expand KOSPI diversification through alternative sectors such as defense. However, his analysis ultimately points to a more fundamental question: when McKinsey predicts that AI data center capacity needs to exceed $8 trillion by 2030, and Gartner forecasts global corporate AI spending will reach $2.6 trillion in 2026, storage chips as a key pillar are simultaneously priced at a discount by investors in three major markets—this contradiction itself may be more noteworthy than any single country's market volatility.

Sources

  • chinatalk.media↗
Continue researching →