Semiconductor plunges 11% in two days: AI computing power bubble burst or golden pit?

On the first two trading days of July 2026, the global semiconductor sector experienced a severe sell-off rarely seen in recent years.

As of the close of U.S. stocks on July 3, Beijing time, the Philadelphia Semiconductor Index (SOX) settled at 12,626.22 points, plunging 5.44% in a single day. The previous trading day, the index had already fallen over 6%. The two-day cumulative drop reached 11%, marking the largest two-day decline in nearly a month.

This sell-off was not an isolated event. Teradyne plummeted 13.63%, KLA fell 11.51%, Sandisk plunged over 14%, retreating about 27% from its recent high and entering bear territory. Arm Holdings dropped 6.58%, Micron Technology fell 5.49%, and AMD declined 4.26%. Even NVIDIA, long considered the leader in AI hardware, did not escape—closing down 1.39% at $194.83.

Goldman Sachs' basket of AI semiconductor stocks suffered heavy losses, recording their worst two-day performance since tariff day; a basket of memory stocks fell over 18% in the past two days, the most severe two-day decline in 12 years.

However, while the chip sector was bleeding, the Dow Jones Industrial Average surged 1.14% to 52,900.07 points, hitting a record closing high. The Nasdaq Composite fell 0.8%, dragged down by chip stocks, while the S&P 500 was nearly flat.

This extreme market divergence pushes a question to all investors: Is this semiconductor crash a cyclical top signal of the AI computing power bubble, or is it a golden buying opportunity worth betting on?

The Trigger of the Rout: How Two News Items Fractured AI Valuation Logic

The direct triggers of this sell-off were two seemingly independent news items pointing in the same direction.

First news: Meta is planning to build an AI cloud computing business, potentially opening its AI models deployed on Meta's infrastructure to external customers in the future, or directly leasing surplus AI computing power. The market immediately interpreted this as a sign of "oversupply of computing power." Meta's internal utilization rate is about 65%, leaving about 35% of idle capacity available for external leasing.

Second news: AI model company Anthropic is discussing cooperation with Samsung Electronics to develop its own AI chip, considering using Samsung's 2nm process for manufacturing.

The common implication of both messages: The AI industry chain is shifting from "unlimited expansion of capital expenditure" to "focusing on capital efficiency and return on investment." The core narrative that drove AI hardware stocks to skyrocket over the past two years—long-term GPU shortage and tech giants continuously raising capital expenditures—is now being reassessed by the market.

What the market is truly trading is not whether "AI demand has peaked," but that the AI industry is moving from "competing on capital spending" to "competing on capital efficiency."

The Bullish Case: Structural Shortages Remain, Cycle Far From Peak

Despite the sharp downturn in market sentiment, some institutions bullish on the semiconductor cycle do not believe the rally has ended.

Nomura Securities, in a report released on July 1, explicitly refuted the "semiconductor peak theory." The report stated that the AI semiconductor cycle is far from peaking, and the second half of 2026 may see an "epic" supply chain mismatch. As cloud providers continue to expand capital expenditure, shortages of advanced packaging, PCB, CCL, and other components will drive price increases and earnings upgrades. The report emphasized that while TSMC is aggressively expanding its wafer-level packaging capacity, the real supply bottleneck will shift to smaller components such as wafer-level substrates (WoS), printed circuit boards (PCB), and copper-clad laminates (CCL).

Nomura's deployment capacity forecast shows: 26.7 GW in 2026, 32.3 GW in 2027, and 22.9 GW in 2028, corresponding to demand for 4 million to 6 million AI chips per year. 2027 is the peak year for deployment, with the AI cycle top delayed until 2028.

Goldman Sachs offered a different bullish perspective from a capital flow angle. Goldman Sachs derivatives expert Brian Garret noted that investors are underweighting U.S. tech stocks, especially the "Magnificent Seven," and favoring the semiconductor sector. The market is shifting from "rewarding companies that spend" to "rewarding companies that profit"—semiconductors, as beneficiaries of capital expenditure, are attracting capital.

Goldman Sachs' data provides strong evidence: The world's first actively managed memory ETF—Roundhill Memory ETF (DRAM)—has risen 141% since its inception in April. The VanEck Semiconductor ETF (SMH) is up 72%, and the iShares Semiconductor ETF (SOXX) is up 99%. The semiconductor industry's revenue is approaching $1 trillion.

Semianalysis, a prominent semiconductor research firm, also published an article on July 3 stating that Meta's computing power procurement has not slowed but is accelerating—in the first half of 2026, Meta has already signed contracts for over 5 GW of computing capacity. If overall computing power were in surplus, Meta would not continue to invest hundreds of billions in infrastructure.

The Bearish Case: Valuation Bubble and Sustainability of Capital Spending

The logic from the bearish side cannot be ignored either.

Morgan Stanley Chief Investment Officer Mike Wilson issued a warning: The price momentum of semiconductor stocks is approaching historical extremes, with trends highly similar to the trajectory of silver stocks earlier this year—which quickly fizzled after a brief frenzy. Wilson compared this volatility to past commodity crashes (such as silver), suggesting that chip stocks may face a boom-and-bust cycle similar to rare earths and gold.

From a valuation perspective, the Philadelphia Semiconductor Index had a 52-week high of 14,655.29 points and a low of 5,418.32 points. The year-to-date gain once exceeded 90%. After such a massive rally, any narrative loosening can trigger sharp valuation contraction.

A more fundamental concern is the sustainability of capital expenditure. The ratio of capital expenditure to operating cash flow for major U.S. cloud providers has approached 94%, meaning their expansion is near the limit of cash flow and heavily reliant on external financing. Institutions like the Bank for International Settlements have warned that if AI investments cannot translate into revenue, cloud service providers will be forced to cut spending, triggering a chain reaction.

The debate over whether "AI computing power has a bubble" is heating up. The current decline in AI hardware appears more like a high-level pullback combined with a loosening of industry narrative: The demand logic has not been disproven, but the pricing assumption that "computing power will never be enough" is being reassessed by the market.

Golden Opportunity or Cycle Peak: Three Scenarios

Based on existing data and logic, three possible scenarios can be inferred:

Scenario 1: Golden Opportunity

The core logic supporting this scenario is that the fundamentals of the semiconductor industry have not fundamentally reversed. AI training computing power remains in short supply. The performance growth of core hardware vendors like NVIDIA remains strong, which is fundamentally different from the "hollow" companies with no revenue and no profits during the 2000 dot-com bubble. If Nomura's warned "epic" supply chain mismatch materializes as expected, price increases and earnings upgrades will drive the industry upward again. Historically, the semiconductor industry often experiences strong rebounds after significant pullbacks—over the past month, semiconductor stocks have risen more than 34%.

Scenario 2: Cycle Peak

The core logic supporting this scenario is that the marginal growth rate of AI capital expenditure is slowing. In 2026, AI investment is still expected to grow 51% year-over-year, but that's a clear slowdown from the 104% growth in 2025. The ratio of cloud providers' capital expenditure to operating cash flow is approaching the 94% limit. Memory chip cyclical peaks are often accompanied by extreme valuation expansion—the DRAM ETF surged 141% in just months since inception, and such speed inherently carries mean-reversion pressure. Institutions like Citron believe the current memory cycle is replicating the pattern of peaks seen in 2008, 2012, and 2018.

Scenario 3: Structural Divergence

This is the most likely scenario. The "rising tide lifts all boats" golden era of semiconductors is over, replaced by extreme divergence. The cycle positions of AI chips (GPU, HBM, advanced packaging) are completely different from traditional chips (PC, smartphone, automotive). The valuation logic for equipment stocks (Teradyne, KLA) differs from design stocks (NVIDIA, AMD). Goldman Sachs' statement that "funds are rotating from the Magnificent Seven to semiconductors" is essentially a structural rebalancing within the industry, not a simple bullish/bearish binary.

Conclusion

As of July 3, 2026, Beijing time, the Philadelphia Semiconductor Index closed at 12,626.22 points, falling 11% over two days. NVIDIA was at $194.83, down 1.39%. Bitcoin rebounded above $61,500.

These three numbers together outline the core contradiction of the current market: Semiconductors are undergoing valuation compression, while crypto is repairing liquidity expectations—the same macro environment, but two very different pricing logics.

The essence of this semiconductor sell-off is a valuation adjustment during the transition of the AI industry from the "capital expenditure expansion phase" to the "capital efficiency verification phase." It is neither a simple "bubble burst" nor a pure "golden opportunity," but a stress test of structural industry change.

For investors, the key is not to judge "top" or "bottom," but to identify divergence: Which companies' valuations have already fully priced in the AI narrative? Which companies still have room for upward earnings revisions? Which parts of the supply chain are seeing narrowing gaps, and where are bottlenecks just beginning to appear?

The long-term story of semiconductors is far from over, but the rhythm of storytelling has changed.

FAQ

Q1: The Philadelphia Semiconductor Index fell 11% in two days. Are declines of this magnitude historically common?

A two-day drop of 11% is a sharp move but not unprecedented. The semiconductor industry has historically been highly volatile. The SOX index has a 52-week low of 5,418 points and a high of 14,655 points, indicating extreme range. The key is to determine the nature of the decline—whether it's a trend reversal or a short-term correction—by combining fundamentals and valuation analysis.

Q2: NVIDIA was relatively resilient in this rout. What does that imply?

NVIDIA fell only 1.39% on July 3, significantly outperforming equipment stocks (Teradyne -13.63%) and memory stocks (Sandisk -14%+). This reflects the market's view that NVIDIA, as a core AI computing supplier, has the strongest competitive moat and highest earnings certainty. However, resilience does not mean safety; if overall AI capital spending slows, NVIDIA will also face the risk of order expectation downgrades.

Q3: Is AI computing power really in oversupply?

There is currently insufficient evidence of a general oversupply. Meta's internal utilization rate is about 65%, with idle capacity, but this is more a resource allocation issue than a demand shortfall. Semianalysis points out that Meta's computing power procurement is still accelerating. The real risk is not "oversupply" but that "investment returns from computing power fall short of expectations," leading to a slowdown in capital expenditure growth.

Q4: Is the crypto market rebound related to the semiconductor sell-off?

There is no direct causal relationship, but they share the same macro backdrop—weak U.S. June nonfarm payroll data lowered interest rate hike expectations. Crypto markets trade on improved liquidity expectations and thus rose, while semiconductors traded on changes in AI capital expenditure logic and thus fell. This divergence precisely illustrates that the current market's main theme is not macro liquidity but structural revaluation of industry fundamentals.

Q5: How to view the semiconductor cycle for the second half of 2026?

Institutional views are divided. Nomura sees an "epic" supply chain mismatch in H2 2026, driving price increases and earnings upgrades. Goldman Sachs sees funds rotating from tech giants to semiconductors. However, some institutions worry that cloud providers' capital expenditure is approaching cash flow limits. Overall, structural divergence is the most likely scenario—AI chip-related segments still have support, while traditional chip segments face more pressure.