The semiconductor index drops into a bear market: has the AI chip cycle peaked? Why did SK Hynix, SanDisk, and Micron all plunge at the same time?

On July 16, Eastern Time, US stocks’ semiconductor sector experienced a systemic selloff. By the close, the Dow Jones Industrial Average fell 0.2% to 52,552.97 points, the S&P 500 dropped 0.51% to 7,533.77 points, and the Nasdaq tumbled 1.47% to 25,881.95 points. The key force dragging the broader market came from chip stocks—a sector whose weight in the S&P 500 has already exceeded 20%.

The Philadelphia Semiconductor Index (SOX) plunged 4.29% in a single day, closing at 11,867 points. That was a cumulative pullback of more than 22% from its historical peak in mid-June, formally putting it into a technical bear market. The Fear and Greed Index also sharply flipped to negative the same day, with market sentiment quickly sliding from optimistic to cautious.

This correction is not a problem with a single company, but rather a repricing of the entire AI hardware value chain. From memory chips to optical communications, from equipment manufacturing to server components, almost no part was spared. Memory and optical communications became the worst-hit areas: SK Hynix ADR crashed 13.48%, SanDisk fell 12.63%, Seagate Technology dropped 10%, and Western Digital slid 9.22%; in the optical communications link, Corning fell more than 9%, Myway Technology (Mairwell?) dropped 8.71%, and Lumentum fell 6.1%. Even Taiwan Semiconductor Manufacturing Co. (TSMC) with strong performance was not immune—second-quarter net profit surged 77% year over year, setting a record high and beating market expectations. Yet at the same time, the company raised its full-year capital expenditure guidance from $52 billion–$56 billion to $60 billion–$64 billion, which instead triggered investor doubts about whether “AI capital expenditures can remain sustainable.” TSMC ADR fell 2.32% that day.

Why memory leaders became the core of the crash

In this selloff, the decline in memory chip companies far exceeded the average level of the semiconductor sector. Behind this were multiple overlapping factors.

SK Hynix ADR closed at $152.31 on Thursday, plunging $24.15 in a single day, down 13.69%, with trading volume of $8.57B. The stock debuted on July 10 at an offering price of $149, and touched a new listing high of $194.80 on July 14. This means that within just two trading days, SK Hynix ADR retreated about 22% from its peak, nearly giving back essentially all of its gains since listing. Its current price is only slightly above the offering price. SK Hynix’s ordinary shares listed in Korea also fell about 11.5% on the same day.

SanDisk closed at $1,411.08, down 12.63%, rewriting the low since May 20 and becoming the largest percentage decliner among S&P 500 constituents that day. Seagate Technology fell to $745.49, down 10%; Western Digital fell to $466.81, down 9.15%; and Micron fell to $853.20, down 5.65%.

Memory became the worst-hit sector, mainly because the market began to reassess two key questions.

First, whether the growth expectations for AI memory have already been priced in excessively. Over the past year, AI server demand surged explosively, making HBM (high-bandwidth memory) the most scarce type of chip, benefiting SK Hynix, Samsung, and Micron deeply. But since 2026, SK Hynix’s Korean-listed shares are up more than 180%, and Micron’s 52-week high has reached $1,255—valuation expansion is already running ahead of fundamentals. When the market starts to question whether AI capital expenditure in 2026 can maintain high-speed growth and whether cloud providers will reduce their GPU purchasing pace, high-valuation stocks naturally become the first targets of selling.

Second, the storage industry’s inherent cyclical characteristics are now coming into play. The storage industry has clearly cyclical attributes: demand growth → supply tightness → price increases → companies expanding capacity, followed by a downward phase of increased capacity → improved supply-demand balance → falling prices. Although AI-related HBM demand remains strong, weakness in the broader DRAM market, NAND market, and consumer electronics demand could drag the overall storage cycle. Analysts noted that TSMC’s second-quarter earnings showed capital expenditure above expectations (above $60 billion), which instead triggered market worries about oversupply in the chip industry.

What the market is worried about: whether AI capital spending is already topping

The core contradiction behind this selloff is not that AI demand has disappeared, but that investors have begun to doubt whether massive AI infrastructure spending can quickly translate into profits.

The logic chain that previously drove semiconductors higher was clear and simple: AI demand growth → GPU demand growth → HBM demand growth → chip company profit growth → stock price rise. But the market’s focus has shifted now: can the growth rate of AI revenue match the scale of data center spending, GPU procurement costs, electricity costs, and the capital expenditures for semiconductor capacity expansion?

TSMC’s second-quarter report offers the latest window. Net profit jumped 77% year over year and beat market expectations—this should have been positive. Yet the market interpreted it as a signal of “good news fully priced in.” At the same time, TSMC raised its full-year capital expenditure guidance to $60 billion–$64 billion. In an upcycle, an expansion signal is viewed as a growth story, but at today’s elevated valuations it is seen as a risk of oversupply.

Bloomberg strategist Tatiana Darie said that the selloff in chip stocks is approaching the technical threshold of prior years’ repeated bottoming-and-bouncing episodes. But whether they can stabilize “still depends on whether ultra-large-scale cloud computing companies continue to raise their expectations for AI capital expenditures.” JPMorgan research shows that over the past five to six weeks, hedge funds have significantly reduced AI-related exposure and holdings in leveraged ETFs. A shift in fund flows further amplifies the magnitude of the sector’s adjustment.

On the macro front, hawkish signals from the Federal Reserve also put pressure on high-valuation tech stocks. Dallas Fed President Lorie Logan (an FOMC voter in 2026) explicitly called for “appropriate” further rate hikes, saying this week’s inflation data is “still not satisfactory”; Kansas City Fed President Schmid also warned that the risk of inflation accelerating again in the coming months remains. CME’s “FedWatch” shows the probability of a rate hike by late July is only 11.2%, but the probability of a cumulative 25 basis point hike by September is already 46.2%. With elevated valuations overlapping tightening expectations, the valuation anchor for tech stocks is loosening.

Does the semiconductor crash mean the AI bubble has burst?

There are significant differences in market views on this issue, and it needs to be examined from both bearish and bullish angles.

The bearish logic is primarily supported by valuation and positioning/position structure. Some AI stocks surged dramatically over the past year, and there is an objective pressure for profit-taking after the run-up. SK Hynix ADR soared from the $149 offering price to $194.80 within just a few days of listing, then plunged back to $152.31 over the next two trading days—this level of volatility itself reflects the fragility of the trade/chip structure. SpaceX’s situation is even more cautionary: according to S3 Partners data, about 185 million shares of SpaceX are currently sold short, representing nearly 29% of the float, corresponding to short positions of about $25 billion. Three weeks ago, this proportion was only 5%–7%. The sharp expansion in short positions indicates that market willingness to short AI-related names is rising rapidly.

When Buffett said in a CNBC interview on July 15 that when everyone is eager to “bet” on AI and other themes, finding truly valuable investments becomes “increasingly difficult,” he compared the current US stock market to a “church with a casino attached.” This warning from a value-investment benchmark further strengthens the market’s questioning of whether AI sector valuations are reasonable.

The bullish logic emphasizes that fundamentals have not undergone a fundamental change. Demand for AI infrastructure is still expanding—cloud companies’ capital expenditures, data center construction, HBM demand, and AI server demand have not shown any material signs of slowing. SK Hynix previously stated that HBM supply had already sold out in advance, and AI storage demand continues to grow. Bloomberg’s Michael Ball believes this selloff is more “mechanical” than driven by fundamentals breaking, meaning the main logic behind AI capital expenditures has not been broken. The record profits and raised forward guidance from TSMC and ASML also support the continued expansion of AI infrastructure buildout.

Yao Yuan, Senior Investment Strategy Officer for Asia at Crédit Agricole Asset Management’s Investment Research Institute, also said that the recent pullback in the Philadelphia Semiconductor Index and the intense volatility in core AI names are more like a short-term technical correction rather than a structural top in the capital expenditure cycle.

Overall, what this looks like is closer to a “valuation correction” rather than a “logic collapse.” The long-term demand trend for AI has not been falsified, but short-term valuation bubbles and crowded positioning have already triggered a systemic adjustment.

What to watch next for semiconductors

Nvidia’s earnings report will be the next key observation point. The market needs to look for clues from data center revenue, GPU demand, and AI customers’ capital expenditure plans to judge whether AI hardware demand is still expanding on its trajectory.

Earnings from SK Hynix and Micron are also crucial. Focus areas include HBM order visibility, the trajectory of DRAM prices, and changes in gross margins—these indicators will directly reflect the true heat of AI storage demand.

Capital expenditures by cloud providers are the most decisive forward-looking indicator. Whether Microsoft, Amazon, Google, and Meta continue to expand their AI infrastructure spending will determine whether the overall sentiment of the AI hardware industry chain can last. Bloomberg strategist’s view is worth referencing: whether chip stocks can stabilize “still depends on whether ultra-large-scale cloud providers continue to raise their expectations for AI capital expenditures.”

In addition, the price trend of storage chips, orders for semiconductor equipment, and changes in global macro interest-rate conditions will all influence the semiconductor sector’s valuation center.

FAQ

Q: What does the Philadelphia Semiconductor Index falling into a technical bear market mean?

A technical bear market usually means the index has fallen by more than 20% from its recent high. This signals that market sentiment is shifting from optimistic to cautious, but it does not necessarily mean the start of a long-term bear market. The current adjustment more reflects profit-taking in high-valuation stocks and a switching of fund style.

Q: Why did SK Hynix ADR plunge nearly 14% in a single day?

Multiple factors in combination: tighter Korean regulation affecting leveraged ETFs triggered a deleveraging chain reaction; the short-term run-up of the AI memory leader was too large—after rising from $149 to $194.80 within days of listing, profit-taking flooded out; doubts emerged about whether AI capital expenditures can be sustained, leading high-valuation names to be sold first.

Q: Has the AI chip cycle really ended?

A more accurate description is a “cycle correction” rather than a “cycle ending.” The long-term demand trend for AI infrastructure has not changed, HBM supply is still tight, but short-term valuation bubbles and crowded positioning have already triggered a systemic adjustment. The subsequent trend will depend on whether cloud providers’ capital expenditures can continue to expand.

Q: Does the crash in storage chip companies reflect worsening fundamentals?

Not entirely. The storage industry has distinct cyclical characteristics, and the market concern now is that capacity expansion could reverse the supply-demand relationship. But AI-related HBM demand remains strong; the weakness in the ordinary DRAM and NAND markets is the factor dragging down the overall storage cycle.