Introduction: Overnight, the index hit a new high, and storage stocks crashed, with both events occurring in the same market and trading day. Many interpretations attribute this to Toshiba's increased production, Seagate's acquisition efforts, and hedging ahead of earnings reports—all of which are true, but none explain why the drop was so severe and why only storage stocks fell. What truly matters is the answer to the question "who is falling, who is rising": the money in AI has begun to change its focus.
I. Let’s State the Facts: Who is Falling, Who is Rising, and What the Index is Doing
On October 6, Eastern Time, the S&P 500 closed up 0.58% at 7,818.93 points, marking its first close above 7,800; the Nasdaq gained 0.45% to close at 27,599.79 points, setting a new closing high for the second consecutive day; Nvidia slightly refreshed its closing record, with a market cap of about $5.79 trillion, briefly exceeding $5.8 trillion during trading (the above data has been verified).
On the same day: Seagate fell 9.18%, Western Digital fell 6.93%, SanDisk fell 2.56%, and Micron fell 1.73%; SK Hynix in Seoul closed down 3.69%. Based on closing prices and market capitalization, the combined market value of Seagate, Western Digital, Micron, and SanDisk evaporated by about $40 billion. On the rising side: Corning rose 6.03%, Marvell rose 5.81%, Lumentum rose 3.82%, Broadcom rose 3.67%, and AMD rose 2.80%.
It is necessary to correct a popular saying: This is not a decline in semiconductors. The SOXX semiconductor ETF closed flat that day, having risen about 3.9% over the previous five trading days, and was only about 1.2% away from a 60-day high. The index did not drop, computing power is rising, and only storage and HDD are collapsing—this structural information is more revealing than the magnitude of the drop: the funds have not left the semiconductor sector but have shifted seats within it.
II. Two Catalysts Cannot Explain This Drop
The market has found two reasons for the decline in storage. One, according to Nikkei news reports, Toshiba plans to invest about 60 billion yen to expand its factory in the Philippines, aiming to double HDD capacity for AI data centers before the fiscal year 2027—this is its first large-scale investment in the HDD business in five years. Second, according to Bloomberg citing informed sources, Seagate and Toshiba are competing to acquire TDK's HDD head business, with the transaction potentially reaching several billion dollars; TDK is the world's only independent supplier of HDD heads.
Both reasons are genuine, but neither can support a single-day drop of 9%. Toshiba's capacity will not double until the fiscal year 2027, which does not address immediate concerns; the potential billions of dollars in spending on the acquisition of the head business does not constitute a fatal variable for Seagate, a company with a market value of $180 billion. What deserves more attention is this layered logic that may affect market pricing.
III. Layered AI Capital Expenditure: The Market is Redefining "Bottlenecks"
First, let’s look at the total amount of money. According to public reports, the combined capital expenditure guidance from Microsoft, Google, Meta, and Amazon for 2026 is about $730 billion, which is about 77% higher than approximately $410 billion for 2025; if Oracle is included, Bank of America estimates that the five companies will invest more than $860 billion this year, approaching $1.2 trillion next year. But on the other hand, the current annual revenue directly attributed to AI services is estimated to be only about $25 billion—about 4% of the capital expenditure. Goldman Sachs' estimates are even stricter: the giants need about $300 billion annually in AI revenue to cover their investments.
As the money increases, the pressure for returns also grows. When capital expenditure shifts from "buying everything" to "targeted breakthroughs," cloud providers' questions change from "how much more to buy" to "which part is choking me."
This layered framework can simultaneously explain every rise and fall on that day:
Bottleneck Layer · Computing Power: Nvidia's GPUs are still in short supply—according to the Financial Times, SpaceX is seeking $40 billion financing led by Apollo Global Management specifically for acquiring Nvidia chips; Marvell raised its fiscal year 2028 revenue guidance to about $20 billion (Wall Street's expectation is about $18.2 billion) during its investor day, and provided a revenue framework of $70-90 billion for fiscal year 2031, far exceeding the market consensus of about $46.85 billion. AMD rose 2.80% and Broadcom rose 3.67% are also included in this layer.
Bottleneck Layer · Optical Interconnection: Upgrading from 800G to 1.6T, Corning rose 6.03% and Lumentum rose 3.82%.
Near Storage Layer · DRAM/HBM: Micron fell only 1.73%—note that Micron's own guidance is that the supply-demand tension in storage will last until 2028, and more than 75% of fiscal year 2027 shipments have been allocated. At least from the disclosures of the above companies, no obvious signs of deterioration have been seen; its smaller decline compared to HDD may be related to its closer alignment with computing power demand.
Peripheral Layer · NAND/HDD Cold Data: The HDD duopoly fell by 6%-9%. Cold data storage is at the outermost layer of the AI training and inference link, making it the first segment to be "negotiated" in cloud providers' budgets. Even in Marvell's own revenue framework for fiscal year 2031, "exchange and storage" is also the smallest segment.
Our view is: the sharp decline in storage is not the market saying "storage is no longer important," but rather that the marginal increment in AI capital expenditure is shifting from "buying everything" to "only buying what is most scarce." Toshiba is expanding HDD rather than DRAM; this detail also offers a perspective for the above framework—its new investment directed towards HDD may reflect the company's judgment regarding long-term demand and supply conditions in that field.
IV. A Bigger Context: The Index’s "Boom" and Individual Stocks’ "Bear Market"
The collapse of storage is not isolated; it occurs in a market with severely deteriorating breadth. According to the latest report from Mike Wilson, Chief Equity Strategist at Morgan Stanley (as reported by the media): 51% of stocks in the Russell 3000 have already declined more than 20% from their June highs, entering a technical bear market, a figure that has worsened from over 40% two weeks ago; the median stock in the S&P 500 is down 16% from its 52-week high, with market breadth at its lowest since the internet bubble burst; there exists a divergence gap of about 12% between the index and breadth.
Wilson's pivot variable is the volatility of U.S. Treasury bonds: Currently, the 10-year U.S. Treasury yield is around 5.3% (verified by market data in this column), the MOVE index is above 100 while the VIX is below 15—if volatility in the bond market cannot cool down, he expects the S&P 500 could drop to about 6,800 to 7,300 points in the next month to complete a "convergence"; otherwise, it will see individual stocks undergo a recovery to fill the gap. This judgment only represents Morgan Stanley’s perspective. Additionally, public statistics show that the weight of semiconductors in the S&P 500 has exceeded 20%, with Nvidia alone accounting for about 13% of the Nasdaq’s weight—this explains why the storage crash could occur on the same day as a new high for a small number of stocks supporting the index while most stocks act independently.
V. What to Watch Next: Three Time Points, Two Switches
The next verification point is tomorrow: On October 8th at 10:00 AM Beijing time, Samsung Electronics will announce its preliminary results for the third quarter. According to Korean media reports, Samsung plans to release preliminary results before the market opens on October 8, with expectations of achieving historical high profitability (FnGuide expectations, cited by the media); the official financial report and earnings call are expected in late October. This is the first verifiable data point following the recent storage sell-off: if Samsung reports preliminary profits below expectations against a backdrop of rising storage prices, the market may be more inclined to interpret it as a "fundamental reassessment"; if preliminary data is strong, the decline in storage that day may be more interpreted as an emotional valuation adjustment.
The second time point: Two formal earnings call meetings in late October. According to media estimates, SK Hynix is tentatively set to announce its official results on October 27, and Samsung is tentatively set for October 29 (neither company has formally confirmed the dates, subject to company announcements). Key points to focus on during the call include not just the quarterly numbers but also management’s wording regarding supply and demand in 2027 and beyond—if "tight" begins to give way to "balance," the market may reassess storage sector valuation frameworks.
The third time point: Cloud vendors' earnings reports from late October to early November. The focus will be on the structure of capital expenditure allocations: whether computing power budgets are maintained, and whether budgets for storage and cold data are being compressed. If structural differentiation is confirmed, "allocation by bottleneck" would shift from inference in this column to fact; if all four types of budgets grow synchronously, the day’s sell-off is more likely to be proven a reversible reallocation.
Two switches lie within the bond market. One is the extreme divergence between the MOVE index and the VIX (MOVE above 100, VIX below 15, according to Morgan Stanley’s report): this is the overall switch for the current fragility of the index; the second is the 10-year U.S. Treasury yield, currently at a high level of around 5.3%, with each additional day spent at this level increasing the difficulty of breadth recovery.
Conclusion
Our judgment has three points. First, what really happened on that day was not a "decline in semiconductors," but a surgical-style reallocation: the SOXX semiconductor ETF closed flat, computing power and optical interconnection rose, while storage and HDD collapsed—indicating that the marginal increase in AI capital expenditure is shifting from "comprehensive expansion" to "allocation by bottleneck," which is our positioning of this storage sell-off. Second, the divergence where HDD fell 9% while Micron only fell 1.7% may reflect that the market is distinguishing between different segments within the "storage" label; Toshiba's investment in expanding HDD rather than DRAM also provides an observational angle for this pricing logic. Third, against the backdrop where 51% of Russell 3000 constituents have entered a bear market, and the index is maintained at new highs by a small number of weighted stocks, any "marginal change" in any segment can be amplified into index-level volatility—monitoring which segments may be further affected by this differentiation is worth continued observation. The above judgments are based on currently available public information, and conclusions may change as conditions evolve.
Data Explanation
U.S. stocks: S&P +0.58% to 7,818.93, Nasdaq +0.45% to 27,599.79; Nvidia +0.14%, Broadcom +3.67%, AMD +2.80%, Marvell +5.81%; storage weakness: Seagate -9.18%, Western Digital -6.93%, Micron -1.73%; SK Hynix -3.69%.
Industry: Toshiba plans to invest 60 billion yen to expand HDD; Seagate and Toshiba are vying to acquire TDK's heads; Marvell’s FY2028 guidance is about $20 billion, and FY2031 framework is $70–90 billion; Micron states supply-demand tightness will last until 2028.
Cloud/AI: The four major cloud companies’ capital expenditures for 2026 are about $730 billion (+77%); SpaceX seeks $40 billion financing to purchase Nvidia chips, according to FT.
Market: 10Y U.S. Treasury around 5.3%; Morgan Stanley states that 51% of Russell 3000 is in a technical bear market, with S&P potentially dropping to 6,800–7,300.
Others: Semiconductors account for over 20% of the S&P 500, and Nvidia accounts for about 13% of the Nasdaq; forward P/E: Seagate 18–20 times, Micron about 8 times, Nvidia about 32 times, AMD about 45 times. Samsung’s preliminary results on 10/8, SK Hynix tentatively on 10/29. Source: exchanges, Nikkei, Bloomberg, FT, etc.; some unverified independently.
Disclaimer: The content of this article is for general information and market commentary only, compiled from publicly available information as of the time mentioned in the text; related market data, expectations, and probabilities may change with market conditions. The views and investment strategies quoted from third-party institutions, analysts, or other individuals represent the viewpoints of those parties at a specific time, and do not represent BIT's views or recommendations. This article does not constitute investment advice, investment research, an offer, solicitation, or endorsement of any security, investment product, or trading strategy, nor should it be used as the basis for any investment decision. Financial markets carry risks, and security prices and market performance may fluctuate; past performance and historical market trends do not represent or guarantee future results. Investors should assess related risks independently based on their own circumstances and seek professional advice when necessary.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
