Global chip stocks plunge amid reassessment of AI investments

Original author: Li Dan

Source: Wall Street Journal

On Tuesday, October 13, the sell-off in AI chip stocks continued to spread, with the global semiconductor sector suffering broad declines from Korea and Japan to the United States.

The Korea Composite Stock Price Index (KOSPI) closed down 5.5%, while Japanese memory chip manufacturer Kioxia fell over 10%. During U.S. trading, the Philadelphia Semiconductor Index (SOX) dropped more than 3% intraday, Micron Technology (MU) slid over 6%, and SanDisk (SNDK), Seagate Technology (STX), and Western Digital (WDC) also plunged, hitting intraday lows of nearly 6% or more.

In the afternoon session, the decline in U.S. semiconductor stocks eased; by close, the Philadelphia Semiconductor Index fell about 0.2%, Micron dropped approximately 3.6%, Western Digital fell about 1.6%, Seagate Technology declined 0.8%, and SanDisk dipped around 0.3%.

This sell-off was not driven by a single factor. Reuters analysis indicates that after nearly four years of enthusiasm for AI infrastructure investment, market sentiment is shifting: rising interest rates, increased volatility in bond markets, higher financing costs for AI, and whether massive capital expenditures will translate into actual returns are all being scrutinized more closely.

Last week, Firmus, a data center operator with NVIDIA's equity stake, put on hold its planned $5 billion IPO in Australia; meanwhile, China's memory chip manufacturer CXMT disclosed progress in developing its 4F² DRAM architecture, further fueling market concerns about the competitive landscape in the storage industry.

Jingxin's technological advancements may serve as a direct catalyst for this round of memory stock sell-offs, but a deeper shift is underway as investors begin to move from chasing AI growth to scrutinizing capital returns.

Wall Street insiders note that among AI-related stocks with significant gains this year, memory chip stocks are particularly vulnerable to profit-taking; in an environment of crowded trades and elevated valuations, new competitive signals could be enough to trigger concentrated selling. However, some analysts argue that ChangXin still lags behind in large-scale mass production, especially in advanced areas such as high-bandwidth memory (HBM), and is not yet capable of posing an immediate, significant threat to Korean manufacturers.

Since OpenAI launched ChatGPT, capital has continued to flow into semiconductor companies, data center operators, telecommunications infrastructure firms, and other businesses benefiting from AI computing power development. This investment boom has driven significant gains in chip stocks and spurred growing capital expenditures on data center construction, servers, and other infrastructure.

But now, the cracks in these transactions are spreading across markets. Reuters noted that signs of adjustment in related assets appeared as early as June, and recently extended to primary and credit markets. Last week, Firmus, a data center operator backed by NVIDIA, shelved its planned $5 billion IPO in Australia due to weakening investor interest, highlighting that capital markets are becoming more cautious in their risk assessment of AI infrastructure projects.

The bond market is also sending cautious signals. As demand for financing related to AI increases, investors are paying closer attention to corporate financing costs and debt repayment capabilities. On Tuesday, Tencent Holdings issued USD-denominated bonds and dim sum bonds in Hong Kong, with the yield guidance for its 10-year USD bond approximately 115 basis points higher than prevailing U.S. Treasury yields. Last month, Japanese tech investment firm SoftBank issued $4.5 billion in high-yield bonds with a seven-and-a-half-year maturity, offering a maximum interest rate of 9.75%.

Interest rates and bond volatility have become key drivers of market repricing. Damian Boy, Portfolio Strategist at Sydney Wilson Asset Management, said the critical shift is not only the rise in bond yields but also increased volatility in the bond market, a change that is gradually influencing the pricing of various assets. He believes the surge in capital spending in the AI sector may be cooling, and some participants could be affected as a result.

BlackRock's research division downgraded its U.S. stock market outlook to "neutral" on Monday, citing potential market volatility from AI infrastructure investments and shifting its preference toward short-term U.S. Treasuries. The International Monetary Fund (IMF) also warned on Tuesday that the impact of AI market adjustments could ripple globally. While AI investments hold promise for economic growth, uncertainty remains over whether these investments will generate returns commensurate with expectations.

This means the market’s focus is shifting: previously, investors primarily bet on sustained growth in computing power demand and related companies increasing capital expenditures; now, capital costs, financing sustainability, and return on investment are becoming key variables influencing valuations.

The decline in Asian markets on Tuesday extended to the U.S. The Korea Composite Stock Price Index (KOSPI) fell more than 6% intraday, marking its fifth consecutive trading day of losses; Kioxia, a memory chip manufacturer listed in Tokyo, closed down over 13%; Samsung Electronics and SK Hynix in South Korea closed down more than 7% and 10%, respectively.

U.S. chip stocks also came under pressure. The Philadelphia Semiconductor Index (SOX) closed down about 2.7% on Monday and fell further during Tuesday’s trading, dropping more than 3% intraday to its lowest level of the day, on track for a fifth consecutive trading day of declines—the longest such streak in nearly 17 months.

Stocks related to storage and data storage saw particularly sharp declines. Micron Technology (MU) fell more than 6% intraday, while Seagate Technology (STX) dropped nearly 8%, Western Digital (WDC) fell over 7%, SK Hynix (SKHY) declined over 6%, and SanDisk (SNDK) slid nearly 6%.

This round of correction has drawn attention because the relevant stocks had already accumulated substantial gains beforehand. As of the close on Monday this week, the Philadelphia Semiconductor Index had still posted a cumulative gain of approximately 72.8% year-to-date, while Micron’s stock had risen over 240% this year. The prior strong rally has made investors more sensitive to valuations, concentration of holdings, and potential competitive risks.

The expanding scope of sell-offs—from South Korean storage giants to U.S. chip stocks—indicates that market concerns are no longer limited to the technological progress of any single company. Investors are reassessing the valuation fundamentals of the entire AI hardware supply chain and whether ongoing increases in capital investment can generate sufficient profits and cash flow.

The development progress of ChangXin's 4F² DRAM architecture has provided a new catalyst for the current correction in memory stocks.

According to Shanghai Securities News, Cao Kanyu, President of CXMT, presented the company's technological explorations in DRAM devices, 4F² architecture, and hybrid bonding at the 4th Integrated Chip and Chiplet Conference held from October 9 to 10, and announced that a new generation of DDR5 RDIMM server memory products integrating these technologies is expected to be launched by the end of this year.

Currently, the DRAM industry standard uses a 6F² cell architecture. CXMT’s proprietary 4F² architecture transforms the memory array layout through its self-developed VSG (Vertical Surrounding Gate) transistor, enhancing gate control, reducing leakage, and shrinking cell size to increase storage density. If this approach can achieve stable mass production, it will provide Chinese manufacturers with a new technological foundation for further entry into the server memory market.

However, the market has no consensus on the extent to which ChangXin's progress will impact existing storage giants.

Robert Pavlik, Senior Portfolio Manager at Dakota Wealth, described this market movement as a wave of profit-taking. He noted that when investors realize potential competition may be on the horizon, market participants may simultaneously rush to exit. This analogy reflects how, in crowded trading sectors, even the mere threat of new competition—before it translates into actual market share shifts—can rapidly trigger concentrated selling.

Media analysis emphasizes that ChangXin's technological advancements are indeed significant, but they should not be directly equated with an immediate threat to existing manufacturers.

Phillip Wool, Chief Market Strategist at Rayliant Global Advisors, believes that Micron’s technological advancements represent a significant moment for China’s semiconductor industry, but the scale of the sell-off in overseas chip stocks on Tuesday may be disproportionate to the actual near-term threat posed. He notes that Micron has yet to prove its ability to achieve large-scale mass production and holds a market share far below that of Samsung Electronics and SK Hynix. More importantly, the current discussions primarily concern conventional DRAM architectures, not HBM; for the foreseeable future, Korean firms still hold a clear advantage in the HBM space.

Vey-Sern Ling, Senior Equity Advisor at UBS, also noted that investors are looking for reasons to take profits, particularly in AI-related stocks and memory stocks that have performed strongly this year. Chinese manufacturers still have a long way to go in advanced areas such as HBM.

These two perspectives are not contradictory: JCET's technological advancements may alter investors' expectations about future competition, but this does not mean it has yet developed the capability to fully challenge international memory giants. For memory stocks that have already experienced significant gains, a shift in competitive expectations may serve as a reason to reduce positions, rather than definitive evidence that current profit fundamentals have reversed.

Elon Musk also commented on the development of China's memory chip industry on social media on Monday. He stated that once China masters the manufacturing method of a certain product, its scale production capacity often surpasses that of other regions worldwide, citing lithium-ion batteries and the solar industry as examples. He also noted that China appears to be rapidly overcoming universal memory specifications such as LP5 and LP6.

However, technological advancements, product launches, customer validation, and large-scale mass production are distinct phases. CXMT’s progress has heightened market awareness of intensifying competition, but the current disclosures are insufficient to demonstrate that it has established a commercial competitiveness in the high-end server memory market capable of transforming the industry landscape.

The correction in chip stocks is just part of a broader shift. As AI infrastructure continues to consume capital, market attention is growing toward the financing costs, investment scale, and future cash flows of related companies.

On Monday, at an event in Singapore, Broadcom CEO Chen Fuyang stated that capital may become one of the biggest constraints on the growth of the AI industry. This remark highlights the practical challenges facing the AI investment boom: even as demand for computing power remains strong, capital supply and funding costs could limit the pace of development.

At the same time, investors must also assess whether investments in AI infrastructure can consistently translate into revenue and profits. Data centers, chips, and power infrastructure require long-term commitments, and whether the associated returns can be realized in a timely manner will impact companies’ ability to raise capital, cash flow, and valuation.

The IMF also warned in its latest report on Tuesday that the economic returns from AI-related investments may fall short of expectations among some market participants. If investments continue to expand without corresponding increases in actual profits and productivity, further repricing of related assets could occur.

Thus, this round of chip stock declines reflects not only concerns about competition in the storage industry, but also increased sensitivity of the entire AI sector to funding costs and return rates.

Despite a sustained sell-off in chip stocks, the market has not broadly concluded that the AI investment cycle has ended. Following the release of a new AI model on Monday, Meta’s stock rose approximately 3.4% against the trend, indicating that investors’ assessments of specific technological advancements and product competitiveness have not fully aligned with the broader semiconductor sector.

Marc Velan, Investment Head at Lucerne Asset Management in Singapore, believes that the sharp volatility in the Korean stock market reflects more about position adjustments and capital cost issues than a deterioration in the fundamental demand for AI. He assesses that the AI cycle may be entering its next phase, with investors increasingly focusing on commercial monetization capabilities and return on invested capital, rather than simply betting on infrastructure expansion.

This also explains why the current sell-off in chip stocks cannot be simply equated with a collapse in AI demand. Whether AI computing demand continues to grow, whether infrastructure investments can generate reasonable returns, and whether competition in the storage industry will pressure profit margins are interrelated but distinct issues.

For the market, the key question going forward is not just how much more AI companies will invest, but whether these investments can be converted into revenue, profits, and cash flow. The stalled star IPOs, rising financing costs, increased bond volatility, and shifting competitive expectations due to ChangXin’s technological progress have collectively shaped the backdrop for this shift in transaction logic.

The AI investment boom may not be over, but the phase of driving valuations solely on growth expectations is now facing stricter scrutiny. As rising capital costs, shifting competitive dynamics, and uncertainty around business returns are all factored into pricing, the recent correction in global semiconductor stocks may be a signal that the market is reassessing the risks and rewards of AI investments.