From KOSPI's single-day drop of 10% to Micron's rise of 17%: The dramatic volatility and restructuring of the AI chip industry narrative within four days.

From June 23 to June 26, 2026, the global semiconductor market experienced a complete narrative collapse and reconstruction. Within four days, the KOSPI index triggered circuit breakers twice, the Philadelphia Semiconductor Index plummeted nearly 8% in a single day, and then a Micron earnings report with a gross margin of 84.9% drove its stock up nearly 16% after hours.

This was not a simple fluctuation of ups and downs, but a market stress test on the sustainability of AI chip demand. When Broadcom failed to raise its full-year AI chip guidance in its early June earnings report, the market began to question the narrative that "computing power is always scarce." When the KOSPI fell from above 9,000 points to a circuit breaker within six trading days, the market began to reprice the supply-demand balance of HBM and AI memory. And when Micron responded to all doubts with $41.46 billion in revenue and an 84.9% gross margin, the market declared with a 17% surge: the capital expenditure cycle for AI chips is far from over.

Taking the event chain timeline as the main thread, we dissect the logic of market sentiment shifts within these 96 hours, and through SMH capital flows, individual stock performance, and industry fundamental data, we evaluate the pricing framework for AI chip stocks in the second half of 2026.

Collapse: The Chain Reaction of Expectation Gaps

Broadcom's "No Upside" Triggers the First Wave of Weakness

On June 3, 2026, after the market close, Broadcom released its Q2 earnings: total revenue of $22.19 billion, up 48% year-over-year, and AI semiconductor revenue of $10.8 billion, up 143% year-over-year. Looking at the numbers alone, this was an excellent report. But the market was focused not on the past, but on the future. Broadcom gave Q3 AI semiconductor revenue guidance of $16 billion, below the analysts' consensus expectation of $17.2 billion. More importantly, management did not raise the full fiscal year 2026 AI chip sales expectation of $56 billion, while the analysts' average expectation for that number was around $57.6 billion.

This is a classic case of an "expectation gap"—the numbers themselves were good enough, but not good enough relative to "the better that had already been priced in." After the earnings release, Broadcom's stock plunged 13% after hours. This decline was not a denial of Broadcom's fundamentals, but the first correction to the market's implicit assumption that "AI chip demand goes infinitely upward."

KOSPI Circuit Breaker: The Multiplicative Effect of Leverage and Panic

On June 23, South Korea's KOSPI index plunged 9.99% to 8,203.84 points, triggering a circuit breaker. Samsung Electronics and SK Hynix both fell more than 12%. The trigger was rumors that SK Hynix might slow down the expansion of HBM4 production, but the deeper reason was the amplifying effect of leverage—South Korea's financial regulator had previously warned that leveraged funds linked to Samsung Electronics and SK Hynix had significantly increased market volatility since their launch.

The KOSPI index was still up about 90% year-to-date in 2026, making it one of the best-performing stock markets globally. This meant that before the crash, Korean chip stocks had accumulated extremely hefty unrealized gains. When a tiny crack appeared in expectations, the domino effect of leveraged funds magnified the sell-off into a circuit breaker.

US Chip Stocks Follow the Decline: Cross-Market Sentiment Contagion

On June 24, the sell-off crossed the Pacific. The Philadelphia Semiconductor Index plunged 7.87% in a single day to close at 13,482 points. SanDisk and Micron Technology fell over 13%, ARM fell over 10%, Qualcomm and Western Digital fell 8%, TSMC and Intel fell over 6%, AMD fell over 5%, and Nvidia fell 4.15%. The Nasdaq fell 2.21%.

This round of selling was not driven by fundamental deterioration of any single company, but by a combination of multiple pressures: rising expectations of a Fed rate hike, investor skepticism about the return on AI capital expenditures by cloud giants, and cross-market contagion of panic from the Korean market. When signals emerged such as "computing power rental prices falling from highs and tech giants collectively tightening AI budgets," the simple narrative of the past year—"the scarcer the computing power, the more reasonable the capital expenditure"—faced its first systemic challenge.

Reconstruction: One Earnings Report Rebuilds All Logic

Micron's "Blockbuster" Moment

On June 24, after the market close, Micron released its fiscal Q3 2026 earnings. The data: revenue of $41.46 billion, up 346% year-over-year, far exceeding analysts' expectations of $35.84 billion; adjusted EPS of $25.11, far exceeding expectations of $20.78; gross margin of 84.9%, up from 74.9% in the previous quarter and 39% in the same period last year.

The 84.9% gross margin ranked highest among large US tech companies, surpassing Meta's most recent quarter's 81.9% and Nvidia's 75%. This was not just about "selling a lot"—it was the ultimate expression of "pricing power." Against the backdrop of persistent HBM supply shortages, whether it's Nvidia, AMD, or Google, all have to accept Micron's pricing.

What excited the market even more was the forecast: Q4 revenue guidance of $49 billion to $51 billion, EPS of $30 to $32, and gross margin heading to 86%. Micron has signed 16 long-term agreements with data center operators, automakers, and other customers, locking in sales for the next three to five years. CEO Sanjay Mehrotra said, "The tight supply situation for memory will continue beyond 2027."

After the earnings release, Micron's stock surged nearly 16% after hours. In regular trading on June 25, MU closed at $1,167.88, up 11.5%. Several investment banks raised their price targets: Needham from $1,550 to $1,650, Bank of America from $1,500 to $1,550, and Susquehanna from $1,750 to $2,000.

Chain Reaction: From Memory to the Entire Industry Chain

Micron's earnings report not only saved its own stock price but also reconstructed the entire narrative framework for AI chips. After the earnings release, US chip stocks all rose after hours: Western Digital rose over 11%, SanDisk over 10%, Qualcomm over 10%, Seagate Technology over 8%, and ARM over 5%.

On June 25, the Philadelphia Semiconductor Index surged 482.68 points (3.59%) to close at 13,940.87. Although the Nasdaq index still fell 0.46% that day to close at 25,358.60, the semiconductor sector had already completed a V-shaped reversal.

SMH Capital Flows: Retail Capital's "Vote with Their Feet"

Capital flows provided another dimension to the narrative reconstruction. In mid-June, the VanEck Semiconductor ETF (SMH) saw a single-day net inflow of approximately $6.9 billion, with AUM rising to $78.9 billion. In late June 2026, US semiconductor ETF retail buying reached about $12 billion, a 1,200% increase from April, a record high. The Roundhill Memory ETF (DRAM) saw its AUM soar to nearly $17 billion within five weeks of listing, the fastest asset accumulation in ETF history.

Massive inflows of retail capital usually indicate that trading has become crowded. But from another perspective, $12 billion in retail buying also shows that after Micron's earnings report verified the sustainability of AI memory demand, retail investors' confidence in the "long-term AI chip narrative" not only did not collapse but found a window to add positions during this sell-off.

Deep Logic of the Narrative Shift

From "Computing Power Scarcity" to "Memory Bottlenecks"

The core shift in this narrative is the market's focus moving from GPU computing power to memory supply. Broadcom's "no upside" shook faith in infinite demand for GPU/ASIC computing power, but Micron's earnings report proved another, harder constraint—the supply shortage of HBM and DRAM is physical, not demand-driven.

Micron's 84.9% gross margin shows that under persistent HBM supply constraints, memory manufacturers have stronger pricing power than chip design companies. This is not about "AI demand slowing down," but about "the value distribution in the AI industry chain rearranging"—tilting from GPU manufacturers to memory suppliers.

Demand Visibility for 2026-2027

Micron's 16 long-term agreements provide direct evidence of demand for 2026-2027. Customers are willing to lock in supply for the next three to five years at current high prices, indicating that downstream expectations for AI memory demand are not weakening but strengthening. This forms an interesting hedge against the suggestion from Broadcom's lack of guidance raise that "demand may fall short of expectations"—different sub-sectors are at different stages of the supply-demand cycle.

Conclusion: The Collapse Is Over, But Volatility Has Not Ended

Within four days, the AI chip narrative completed a full stress test. Broadcom's "no upside" exposed the excessive optimism implicit in market pricing; the KOSPI circuit breaker revealed the fragility of leveraged capital; and Micron's 84.9% gross margin used irrefutable data to prove the realism and sustainability of AI memory demand.

As of June 26, Bitcoin remains in a weak consolidation range of $58,000-$60,000, with the Fear and Greed Index at 13, indicating extreme fear. Risk appetite in the crypto market and semiconductor market has not fully recovered. However, SMH's single-day inflow of $6.9 billion and DRAM ETF's $17 billion asset accumulation over five weeks indicate that capital has not left the AI chip track; it is just waiting for a more certain pricing anchor.

Micron's earnings report provided that anchor. In the second half of 2026, the pricing logic for AI chip stocks will shift from "dreaming of infinite demand" to "verifying real supply constraints"—the former relies on imagination, the latter on gross margins and long-term agreements. For investors, this means a more quantifiable valuation framework and also more dramatic stock divergence.

FAQ

Q1: Why did Broadcom's stock plunge despite beating earnings expectations?

Broadcom's Q2 revenue of $22.19 billion and AI semiconductor revenue of $10.8 billion both beat expectations, but the market focused on increments rather than the base. Q3 AI semiconductor guidance of $16 billion missed the expected $17.2 billion, and the full-year sales target of $56 billion was not raised. With AI chip stocks already valued at extremely high levels, "meeting expectations" equaled "below implied expectations," triggering the sell-off.

Q2: What is the transmission mechanism between the KOSPI plunge and the US chip stock sell-off?

The KOSPI plunged 10% on June 23, led by chip giants like Samsung and SK Hynix, triggering a circuit breaker. Since Korean chip stocks are a core part of the global semiconductor supply chain, their crash quickly sparked concerns about global AI chip demand. On June 24, the Philadelphia Semiconductor Index followed with a 7.87% decline, forming a "Korea → Global" panic transmission.

Q3: What does Micron's 84.9% gross margin mean?

An 84.9% gross margin ranks highest among large US tech companies, surpassing Meta (81.9%) and Nvidia (75%). This means that against the backdrop of persistent HBM and DRAM supply shortages, memory manufacturers have extremely strong pricing power. Whether it's Nvidia, AMD, or Google's AI processors, they all rely on Micron's high-bandwidth memory.

Q4: Does SMH's single-day inflow of $6.9 billion mean the semiconductor sector is already overcrowded?

SMH had a single-day net inflow of approximately $6.9 billion, with AUM rising to $78.9 billion. In late June, US semiconductor ETF retail buying reached about $12 billion, a 1,200% increase from April. From a behavioral finance perspective, such concentrated capital inflows do indicate that trading has become crowded, increasing short-term volatility risk. However, crowding itself is not a sell signal—the key is whether fundamentals can continue to validate high valuations.

Q5: How will the investment logic for AI chip stocks change in the second half of 2026?

In the first half, the pricing logic for AI chip stocks was "computing power is always scarce," relying on imagination about long-term demand. In the second half, it will shift to "verifying real supply constraints"—the market will focus more on quantifiable metrics such as gross margins, long-term agreement coverage, and capacity expansion pace. Supply-demand cycles in different sub-sectors (GPU/ASIC vs. HBM/DRAM) will diverge, and individual stock performance differences may widen.