In the first two trading days of July 2026, the global semiconductor sector experienced one of its most dramatic shakeups in recent years.
As of the close of the US stock market on July 3 (Beijing time), the Philadelphia Semiconductor Index (SOX) ended at 12,626.22 points, plunging 5.44% in a single day. The previous trading day, the index had already dropped over 6%. The two-day cumulative decline reached 11%, marking the largest two-day drop in nearly a month.
This crash was not an isolated event. Teradyne tumbled 13.63%, KLA fell 11.51%, and Sandisk plummeted over 14%, retreating about 27% from its recent high and entering bear market territory. Arm Holdings dropped 6.58%, Micron fell 5.49%, and AMD slid 4.26%. Even NVIDIA, long regarded as the leader in AI hardware, was not spared—it closed down 1.39% at $194.83.
Goldman Sachs’ basket of AI semiconductor stocks suffered heavy losses, posting its worst two-day performance since tariff day. A basket of memory stocks fell more than 18% over the past two days, marking the steepest two-day drop in 12 years.
Yet, while the chip sector bled, the Dow Jones Industrial Average surged 1.14% to a record close of 52,900.07. The Nasdaq slipped 0.8% under the weight of chip stocks, while the S&P 500 finished nearly flat.
This extreme market divergence has forced all investors to confront a key question: Is this semiconductor crash a sign that the AI compute bubble has burst at the cycle’s peak, or is it a golden buying opportunity?
The Catalyst: How Two Headlines Shattered AI Valuation Logic
The immediate trigger for this sell-off was two seemingly unrelated news items, both pointing in the same direction.
First headline: Meta is planning to build an AI cloud computing business and may open up its AI models, deployed on Meta’s infrastructure, to external clients or directly lease out surplus AI computing power. The market quickly read this as a signal of "excess compute capacity." Meta’s internal compute utilization is about 65%, meaning roughly 35% of capacity is idle and could be rented out.
Second headline: AI foundation model company Anthropic is in talks with Samsung Electronics to jointly develop custom AI chips, potentially using Samsung’s 2nm process for manufacturing.
Both headlines point to the same trend: the AI industry is shifting from "unlimited capital expenditure expansion" to "focus on capital efficiency and investment returns." The core narrative that has fueled the AI hardware rally for the past two years—persistent GPU shortages and tech giants continually raising capex—is now being re-examined by the market.
The market isn’t really trading on whether "AI demand has peaked," but on the fact that the AI industry is moving from a "capex race" to a new phase focused on "capital efficiency."
Bull Case: Structural Shortages Persist, Cycle Nowhere Near Its Peak
Despite the sharp shift in sentiment, institutions bullish on the semiconductor cycle do not believe the rally is over.
Nomura took a clear stance against the "semiconductor peak" narrative in its July 1 report. The report argued that the AI semiconductor cycle is far from peaking and that the second half of 2026 could see an "epic" supply chain mismatch. As cloud providers continue to ramp up capital expenditures, shortages in advanced packaging, PCBs, CCLs, and other components will drive price hikes and upward earnings revisions. The report emphasized that while TSMC is aggressively expanding its wafer-level packaging capacity, the real supply bottleneck will shift to wafer-level substrates (WoS), printed circuit boards (PCBs), and copper-clad laminates (CCLs).
Nomura’s tracked deployment capacity forecasts: 26.7GW in 2026, 32.3GW in 2027, and 22.9GW in 2028, translating to annual demand for 4–6 million AI chips. The deployment peak is projected for 2027, pushing the AI cycle top out to 2028.
Goldman Sachs offers a different bullish perspective, focusing on capital flows. Goldman derivatives specialist Brian Garret noted that investors are underweighting US tech stocks—especially the "Magnificent Seven"—and shifting into semiconductors. The market is moving from "rewarding companies that spend" to "rewarding companies that earn"—and as beneficiaries of capex, semiconductor firms are now favored by capital.
Goldman’s data backs this up: since its launch in April, the world’s first actively managed memory ETF—Roundhill Memory ETF (DRAM)—has surged 141%. The VanEck Semiconductor ETF (SMH) is up 72%, and the iShares Semiconductor ETF (SOXX) is up 99%. Industry-wide semiconductor revenue is approaching $1 trillion.
Leading semiconductor research firm Semianalysis also noted on July 3 that Meta’s compute procurement hasn’t slowed—in fact, it’s accelerating. In the first half of 2026, Meta has already secured over 5GW of compute capacity. If there were a true compute glut, Meta wouldn’t be ramping up its infrastructure spending by tens of billions.
Bear Case: Valuation Bubble and Capex Sustainability in Question
The bearish logic is equally compelling.
Morgan Stanley CIO Mike Wilson sounded the alarm: semiconductor stock price momentum is nearing historic extremes, closely mirroring the trajectory of silver stocks earlier this year—which fizzled out after a brief frenzy. Wilson compared this volatility to past commodity crashes (like silver), suggesting chip stocks may face boom-and-bust cycles similar to rare earths and gold.
From a valuation standpoint, the SOX’s 52-week high is 14,655.29, and its low is 5,418.32. Year-to-date gains had at one point exceeded 90%. After such a massive run-up, any narrative shift could trigger a sharp valuation contraction.
The more fundamental concern is capex sustainability. The capex-to-operating-cash-flow ratio for leading US cloud providers is now close to 94%, meaning expansion is pushing the limits of cash flow and is highly dependent on external financing. Institutions like the Bank for International Settlements have warned that if AI investments fail to translate into revenue, cloud service providers will be forced to cut spending, setting off a chain reaction.
Debate over whether there’s an "AI compute bubble" is heating up. The current AI hardware pullback looks more like a high-level correction compounded by narrative fatigue: the demand story hasn’t been disproven, but the pricing assumption of "compute is always scarce" is being re-examined by the market.
Golden Opportunity or Cycle Top: Three Scenario Projections
Based on current data and logic, three possible scenarios emerge:
Scenario 1: Golden Opportunity
The core logic here is that the semiconductor industry’s fundamentals haven’t undergone a structural reversal. AI training compute remains in short supply. Core hardware players like NVIDIA continue to post robust earnings growth, fundamentally different from the "hollow" dot-com era of 2000 when many firms had no revenue or profit. If Nomura’s warning of an "epic" supply chain mismatch materializes, price hikes and earnings upgrades will drive another industry upswing. Historically, the semiconductor sector often rebounds strongly after major pullbacks—over the past month, the sector has risen more than 34%.
Scenario 2: Cycle Top
This scenario is based on the logic that marginal growth in AI capex is slowing. AI investment in 2026 is still projected to grow 51% year-over-year, but that’s a sharp slowdown from 104% in 2025. The capex-to-operating-cash-flow ratio for cloud providers is nearing the 94% threshold. Memory chip cycles typically peak alongside extreme valuation expansions—DRAM ETF, for instance, has soared 141% in just a few months, a pace that implies mean reversion risk. Firms like Citron argue that the current memory cycle is mirroring the peaks seen in 2008, 2012, and 2018.
Scenario 3: Structural Divergence
This is the most likely scenario. The "rising tide lifts all boats" era for semiconductors is over, replaced by sharp internal divergence. AI chips (GPUs, HBM, advanced packaging) and traditional chips (PCs, smartphones, autos) are at very different points in their respective cycles. Equipment makers (Teradyne, KLA) and design firms (NVIDIA, AMD) also have distinct valuation logics. Goldman’s observation that "capital is rotating from the Magnificent Seven to semiconductors" is really about structural rebalancing within the sector, not a simple bull-bear dichotomy.
Conclusion
As of July 3, 2026 (Beijing time), the Philadelphia Semiconductor Index closed at 12,626.22, down 11% over two days. NVIDIA finished at $194.83, down 1.39%. Bitcoin rebounded above $61,500.
Taken together, these three numbers highlight the market’s core contradiction: semiconductors are undergoing valuation compression, while crypto is recovering on liquidity expectations—two entirely different pricing logics in the same macro environment.
The essence of this semiconductor crash is a valuation adjustment as the AI industry transitions from a "capex expansion phase" to a "capital efficiency validation phase." It’s not simply a "bubble burst," nor is it purely a "golden opportunity"—it’s a stress test for structural change in the industry.
For investors, the key isn’t to call the "top" or the "bottom," but to identify divergence: Which companies’ valuations have fully priced in the AI narrative, and which still have room for upward earnings revisions? Where are supply-demand gaps narrowing, and where are new bottlenecks just beginning to emerge?
The long-term story for semiconductors is far from over—but the pace and rhythm of that story have clearly changed.
FAQ
Q1: Is an 11% two-day drop in the Philadelphia Semiconductor Index historically common?
An 11% drop over two days is highly volatile, but not unprecedented. The semiconductor industry has always been characterized by high volatility—the SOX’s 52-week range is from 5,418 to 14,655, an enormous swing. The key is to determine the nature of the decline—is it a trend reversal or just a short-term correction? That requires a combined assessment of fundamentals and valuations.
Q2: What does NVIDIA’s relative resilience in this crash mean?
On July 3, NVIDIA fell just 1.39%, significantly outperforming equipment stocks (Teradyne down 13.63%) and memory stocks (Sandisk down over 14%). This reflects the market’s view that NVIDIA, as the core supplier of AI compute, has the strongest competitive moat and the highest earnings visibility. But resilience doesn’t mean immunity—if overall AI capex slows, NVIDIA could still face downward revisions to order expectations.
Q3: Is there really an AI compute glut?
There’s no solid evidence that AI compute is broadly oversupplied. Meta’s internal compute utilization is about 65%, with some idle capacity, but this is more a matter of resource allocation than weak demand. Semianalysis notes that Meta’s compute procurement is still accelerating. The real risk isn’t "compute oversupply," but rather that "returns on compute investment fall short of expectations," leading to slower capex growth.
Q4: Is the crypto market rebound related to the semiconductor crash?
There’s no direct causal link, but both share the same macro backdrop—weak US June nonfarm payroll data has lowered rate hike expectations. Crypto is rising on improved liquidity outlook, while semiconductors are falling as the AI capex narrative shifts. Their divergence underscores that the current market’s main theme is not macro liquidity, but a structural revaluation of industry fundamentals.
Q5: What’s the outlook for the semiconductor cycle in the second half of 2026?
Institutions are divided. Nomura expects an "epic" supply chain mismatch in the second half of 2026, driving price hikes and earnings upgrades. Goldman sees capital rotating from tech giants into semiconductors. But some worry that cloud providers’ capex is nearing cash flow limits. Overall, structural divergence is the most likely scenario—AI chip segments remain supported, while traditional chip sectors face greater pressure.
