Global Chip Stocks Slide as Market Reassesses AI Spending

The selloff in AI-related chip names spread across the globe on Tuesday, hitting semiconductor shares from Korea and Japan to the United States. Korea's KOSPI closed down 5.5%, and Japanese memory maker Kioxia lost more than 10%. In U.S. trading, the Philadelphia Semiconductor Index (SOX) fell more than 3% intraday; Micron (MU) slid over 6%, while SanDisk (SNDK), Seagate (STX) and Western Digital (WDC) each touched intraday lows of roughly 6% or worse. Losses narrowed in the afternoon: the SOX finished about 0.2% lower, Micron fell roughly 3.6%, Western Digital about 1.6%, Seagate 0.8% and SanDisk around 0.3%. No single trigger explains the move. Reuters notes that after nearly four years of enthusiasm for AI infrastructure, sentiment is turning. Higher interest rates, a more volatile bond market, rising AI financing costs and doubts over whether huge capital spending will translate into returns are all under scrutiny. Two company-specific developments added to the pressure: Firmus, a data-center operator backed by Nvidia, shelved a planned $5 billion IPO in Australia, and China's CXMT disclosed progress on a 4F² DRAM architecture. The AI trade shifts: from chasing growth to demanding returns Since OpenAI released ChatGPT, capital has poured into chipmakers, data-center operators, telecom-infrastructure firms and other beneficiaries of AI computing power. The boom lifted chip stocks and fueled rising spending on data centers and servers. The cracks are now spreading. Reuters notes that signs of stress in related assets appeared as early as June and have recently reached primary and credit markets. Firmus, the Nvidia-backed data-center operator, put its $5 billion Australian listing on hold last week as investor interest weakened, a sign that capital markets are reassessing AI infrastructure risk. The bond market is flashing caution as well. Tencent Holdings sold dollar bonds and dim sum bonds in Hong Kong on Tuesday, guiding its 10-year dollar note at about 115 basis points over prevailing Treasury yields. SoftBank issued $4.5 billion of seven-and-a-half-year high-yield bonds last month at a maximum interest rate of 9.75%. Rates and bond volatility have become key drivers of repricing. Damian Boy, portfolio strategist at Wilson Asset Management in Sydney, said the critical shift is not only higher bond yields but greater volatility in the bond market, a change that is feeding through into the pricing of assets. He believes the AI capital-spending surge may be cooling and that some participants could be hurt. BlackRock's research arm cut its U.S. equity view to neutral on Monday, citing potential volatility from AI infrastructure investment and shifting its preference toward short-term Treasuries. The International Monetary Fund warned Tuesday that an AI market correction could ripple across the world; the investments hold promise for growth, but it is uncertain whether they will produce the returns investors expect. The market's focus is moving. Investors once bet on sustained growth in computing demand and rising corporate capex; now capital costs, financing sustainability and return on investment are becoming the key variables in valuation. Cross-market selling pressure, from Seoul to New York Asia's declines on Tuesday carried into the United States. The KOSPI fell more than 6% intraday, a fifth straight losing session. Tokyo-listed Kioxia closed down more than 13%. Samsung Electronics and SK Hynix finished more than 7% and 10% lower. U.S. chip stocks also came under pressure: the SOX closed down about 2.7% on Monday and fell more than 3% intraday on Tuesday, heading for a fifth consecutive decline, the longest streak in nearly 17 months. Storage names were hit hardest. Micron fell more than 6% intraday, Seagate nearly 8%, Western Digital more than 7%, SK Hynix more than 6% and SanDisk nearly 6%. The correction stands out because the stocks had already run so far. At Monday's close, the SOX was still up about 72.8% year to date, and Micron had gained more than 240% this year. Strong prior gains have made investors more sensitive to valuations, crowded positioning and competitive risk. With selling spreading from Korean memory giants to U.S. chipmakers, the worry is no longer confined to any single company's technology roadmap. Investors are re-examining the valuation case for the entire AI hardware supply chain and whether rising capital investment can generate sufficient profit and cash flow. CXMT breakthrough stirs competition fears; Wall Street calls it profit-taking CXMT's work on a 4F² DRAM architecture has given the memory selloff a fresh catalyst. At the fourth Integrated Chip and Chiplet Conference on Oct. 9-10, CXMT president Cao Kanyu presented the company's explorations in DRAM devices, the 4F² architecture and hybrid bonding, and said a new generation of DDR5 RDIMM server memory products integrating the technologies is expected to launch by the end of this year. The industry standard today is a 6F² cell. CXMT's proprietary 4F² layout relies on a self-developed VSG (vertical surrounding gate) transistor to improve gate control, reduce leakage and shrink cell size for higher storage density. Stable mass production would give Chinese manufacturers a new technological footing for entering the server memory market. How much that would hurt incumbents is unresolved. Robert Pavlik, senior portfolio manager at Dakota Wealth, described the move as a wave of profit-taking: once investors realize competition may be coming, market participants can rush for the exit at the same time. In crowded trades, even the threat of new competition, before it turns into actual share shifts, can trigger concentrated selling. Analysts caution against equating technical milestones with an imminent threat. Phillip Wool, chief market strategist at Rayliant Global Advisors, said CXMT's advance is a significant moment for China's semiconductor industry, but the scale of Tuesday's overseas selloff may be out of proportion to the near-term danger. CXMT has yet to prove it can mass-produce at scale and holds a market share far below Samsung Electronics and SK Hynix. The discussion centers on conventional DRAM, not HBM, where Korean firms retain a clear advantage for the foreseeable future. Vey-Sern Ling, senior equity advisor at UBS, said investors are looking for reasons to take profits, particularly in AI-related and memory stocks that have performed strongly this year, and Chinese manufacturers still have far to go in advanced areas such as HBM. The two views are not contradictory. CXMT's progress may change expectations about future competition without meaning it can yet challenge international memory giants. For stocks that have already gained sharply, shifting competitive expectations are a reason to trim positions rather than proof that profit fundamentals have reversed. Elon Musk weighed in Monday, saying that once China masters the manufacturing method for a product, its capacity at scale often surpasses that of other regions, citing lithium-ion batteries and solar. China, he added, appears to be rapidly closing on mainstream memory specifications such as LP5 and LP6. Technical progress, product launch, customer validation and large-scale mass production are distinct stages, though. CXMT's disclosures have raised awareness of intensifying competition but do not yet show commercial competitiveness in high-end server memory strong enough to reshape the industry. Capital costs and AI returns become the market's new question The chip selloff is only part of a broader shift. As AI infrastructure keeps consuming capital, attention is turning to the financing costs, investment scale and future cash flow of the companies involved. Broadcom CEO Hock Tan said Monday at an event in Singapore that capital may become one of the biggest constraints on the growth of the AI industry, a reminder that even with strong demand for computing power, funding supply and cost can limit the pace of development. Investors must also judge whether AI infrastructure spending turns consistently into revenue and profit. Data centers, chips and power infrastructure require long-term commitments, and the timing of returns affects companies' ability to raise money, their cash flow and their valuations. The IMF warned Tuesday that the economic returns on AI-related investment may fall short of what some market participants expect; if spending keeps expanding without matching gains in profit and productivity, further repricing of related assets is possible. The decline in chip stocks reflects not only competition worries in memory but also the sector's growing sensitivity to funding costs and returns. The AI cycle is not over; the market is asking who can deliver Despite the sustained selloff, there is no broad conclusion that the AI investment cycle has ended. Meta rose about 3.4% on Monday after releasing a new AI model, showing that views on specific technologies and product competitiveness have not moved in lockstep with the wider semiconductor sector. Marc Velan, head of investment at Lucerne Asset Management in Singapore, said the violent swings in Korean equities reflect position adjustments and capital-cost issues more than any deterioration in underlying AI demand. He sees the AI cycle entering a next phase in which investors focus on commercialization and return on invested capital rather than simply betting on infrastructure expansion. That is why the current selloff cannot be read simply as collapsing AI demand. Whether computing demand keeps growing, whether infrastructure investment earns a reasonable return and whether memory competition pressures margins are related but separate questions. For the market, the key question is not only how much more AI companies will invest, but whether that investment converts into revenue, profit and cash flow. Stalled star IPOs, rising financing costs, a more volatile bond market and shifting competitive expectations after CXMT's progress together form the backdrop to this change in trading logic. The AI investment boom may not be over, but the phase in which growth expectations alone drove valuations is now facing stricter scrutiny. With capital costs rising, competitive dynamics shifting and business returns uncertain, the recent correction in global semiconductor stocks may signal that the market is repricing the risks and rewards of AI investment.