The Semiconductor Correction of July 2026: Anatomy of a Bifurcated Selloff

The semiconductor sector has just undergone its most significant correction of 2026. Over eight trading sessions ending July 2, the Philadelphia Semiconductor Index (SOX) fell 13.7% from its all-time closing high of 14,655 on June 22, landing at 12,626. The VanEck Semiconductor ETF (SMH) posted its best quarter ever in Q2 — up 71% — then gave back 7.0% in the first week of Q3. The SOX index fell 9.4% over the same span.

But headline numbers obscure the real story. This correction is not uniform. It is bifurcated — splitting the semiconductor value chain into two camps with dramatically different trajectories. Understanding that split is the difference between catching a knife and catching the next leg up.

Here is what happened, why, and what to expect when markets reopen July 6.

Anatomy of the Correction

The selloff unfolded in two distinct waves, each with different triggers and different winners and losers.

Wave 1: June 3 — The Broadcom Trigger

The first tremor came on June 3, when a confluence of profit-taking and position unwinding hit a sector that had run too far, too fast. The result was a sector-wide repricing that erased roughly $1.4 trillion in market value across the AI chip ecosystem in a single session. Broadcom dropped 12.6%, Marvell plunged 17%, and the contagion spread to equipment and memory names.

Crucially, the market recovered from this wave. By June 22, the SOX had climbed to a new all-time high of 14,655 — making the early-June dip a buying opportunity for those who took it.

Wave 2: June 23 – July 2 — Profit-Taking Meets Structural Fear

The second wave is the one we're in now, and it is more complex. It began with late-quarter rebalancing — natural after an 87% year-to-date surge in the SOX — but was supercharged by a specific catalyst on July 1: Meta announced it would begin renting out excess AI computing capacity to enterprise customers.

That single announcement did two things. First, it sent Meta shares up 9% (the company transitioning from pure AI capex consumer to AI infrastructure provider). Second, and far more importantly for chip investors, it planted the first credible seed of doubt about whether AI compute supply might be catching up to demand. If hyperscalers have enough capacity to start selling the surplus, the narrative of insatiable, bottomless AI chip demand begins to crack.

July 2 then delivered the second blow: semiconductor equipment and memory stocks collapsed. Memory maker SanDisk fell 14.3%, while equipment companies Teradyne dropped 13.7%, KLA 11.6%, and Lam Research 10.2% — all in a single session where the Nasdaq dropped less than 1%. This was not a market-wide panic. It was a precision strike on the capital-intensive segments of the supply chain.

The Numbers

Note the dispersion. ARM, the most richly valued chip name, is down nearly 25% in two weeks. Qualcomm, down 29.5% from its high, has been punished on a combination of declining revenue and sector-wide de-risking. Meanwhile TSM and AMD are essentially flat over two weeks — the AI-foundry and AI-CPU narratives are holding. This dispersion is the key signal.

The Structural Divide: Designers vs. Equipment Makers

The most important insight from this correction is the widening gap between two segments of the semiconductor value chain that used to move in lockstep.

AI chip designers and foundries — Nvidia, AMD, TSMC — have held relatively well. These companies sell into the AI accelerator shortage, where demand genuinely exceeds supply and pricing power is extraordinary. Micron's recent earnings illustrated this perfectly: revenue more than quadrupled year-over-year, and gross margins exploded from 39% to 84.9% — more than doubling in four quarters. When a company's gross margin more than doubles in a year, demand is not the problem.

Semiconductor equipment and memory manufacturers — KLA, Lam Research, Applied Materials, Teradyne, SanDisk — have been crushed. Equipment companies sell the machines that fabs use to build chips, and their revenue depends on capital expenditure decisions made 12 to 24 months in advance. The July 2 collapse was triggered by a major memory manufacturer signaling NAND oversupply conditions extending through at least Q3 2026, which hit SanDisk (heavily NAND-exposed) the hardest. Equipment names followed as the market repriced the implications for fab capex spending.

The divergence between AI designer/foundry names and equipment/memory makers has reached approximately 35 percentage points year-to-date. This is not random. It reflects a structural reality: AI chip production is concentrated at TSMC's most advanced nodes, benefiting a narrow slice of the equipment ecosystem while leaving the rest dependent on a broad-based fab investment cycle that remains under pressure.

Money Morning summarized the dynamic on July 3: "The equipment sector is essentially a leading indicator for the broader semiconductor cycle. When KLA and Lam sell off this hard on no company-specific news, the market is telling you something about capex intentions 4 to 6 quarters out."

This is the tension at the heart of the correction: AI is booming, but it is a narrow boom. The question is whether that narrow boom can pull the rest of the semiconductor industry up with it, or whether the broader cycle's weakness eventually drags even the AI pure-plays down.

The Macro Backdrop: Why This Isn't 2000 (Yet)

JPMorgan's Jason Hunter drew explicit parallels to the dot-com era in a July 1 client note, pointing to the divergence between AI hardware suppliers (soaring) and hyperscaler stocks (struggling — Meta down 5% YTD, Microsoft down 18% YTD with its worst monthly loss since June 2000). The comparison is worth taking seriously: in 1999, telecom equipment suppliers saw a "parabolic rise" while companies making heavy capital investments in the space tumbled. The dot-com bubble burst less than a year later.

But the macro backdrop in July 2026 differs from 1999-2000 in several critical ways:

Interest rates are falling, not rising. The 10-year Treasury yield sits at 4.37%, down about 11 basis points over the past month and trending lower. In 1999-2000, the Fed was aggressively tightening, with the 10-year yield climbing above 6.5%. Falling yields are a tailwind for growth stock valuations, not a headwind.

The VIX is calm. At 16.15, the VIX is below its 20-day and 50-day averages, sitting in the lower half of its 90-day range (15.32–25.78). There is no fear spike accompanying this correction. This is orderly profit-taking, not a deleveraging event. The VIX actually fell 2.7% on July 2 even as semis sold off — a clear signal that the broader market is not in distress.

Oil is crashing. Crude has fallen 28% in a month to $68.78. This is a powerful deflationary signal that gives the Fed maximum flexibility to cut rates if needed. In 2000, surging energy prices compounded the inflation problem.

The broader market is diverging, not collapsing. The S&P 500 is down just 1.7% over the past month and the Dow hit a record intraday high of 52,742 on July 1. The Nasdaq has been hit harder at -4.7% over the month, but that is a measured tech pullback, not a systemic break. This correction is semiconductor-specific, not market-wide — the exact opposite of the dot-com crash, which was led by large-cap tech.

AI capex is real and enormous. Combined AI infrastructure spending from Meta, Microsoft, Amazon, and Alphabet is on track for $725 billion in 2026. In 1999, the spending was speculative — infrastructure being built for business models that didn't exist yet. In 2026, AI is generating real revenue, even if the ROI debate is legitimate.

The dot-com parallel is a warning about positioning and sentiment, not a forecast. The structural differences matter.

Valuation Analysis: Where Value Meets Risk

The correction has created a wide valuation dispersion that separates defensible positions from vulnerable ones.

The PEG ratio (P/E divided by earnings growth rate) is the most useful metric here. Anything below 1.0 suggests a stock is undervalued relative to its growth trajectory:

• Micron (PEG 0.14) is pricing in an apocalypse that its fundamentals don't support. Revenue grew 345% and gross margins are at 84.9%. Even after the memory cycle eventually normalizes, the forward P/E of 6.5 provides enormous margin of safety. This is the most attractive risk-reward in the sector.

• Nvidia (PEG 0.60) trades at a forward P/E of 15.3 despite 85% revenue growth and 63% profit margins. The market is pricing in a significant growth deceleration. If that deceleration doesn't materialize — and the AI accelerator shortage suggests it won't near-term — the stock is genuinely cheap here.

• Broadcom (PEG 0.41) is in a similar position. Its custom AI accelerator business and networking dominance make it the most diversified AI beneficiary outside Nvidia.

• ARM (PEG 2.60) and ASML (PEG 2.65) are the most vulnerable. ARM trades at 102 times forward earnings with only 20% revenue growth. ASML at 36 times forward with 13% growth. These are the names where any disappointment will be punished hardest, and where the correction has the furthest to run.

• Qualcomm looks cheap on P/E but has negative revenue growth. The market is correctly pricing in structural decline in its handset business despite AI PC and automotive optionality. This is a potential value trap.

Technical Setup: What the Charts Say

The technical picture is nuanced and does not present a clean "buy" or "sell" signal. All indicators below are calculated from standard 14-period RSI, 20-period moving averages, and 20-period Bollinger Bands using daily data through July 2. Here is the honest read:

Moving averages: Every major name except AMD and INTC is trading below its 20-day moving average, confirming short-term downtrend. However, most remain above their 50-day moving averages (QCOM and AVGO are the exceptions, having broken below). This pattern — below MA20 but above MA50 — is characteristic of a correction within a larger uptrend, not a trend reversal.

RSI readings: No stock is in oversold territory (<30). Readings range from 37 (QCOM, the weakest) to 55 (AMD, the strongest). The absence of oversold readings means there is no mechanical bounce trigger. The market could continue lower before technical buyers are forced to step in.

Bollinger Bands: NVDA at $194.83 is testing its lower Bollinger Band ($189.89). The SOX at 12,626 is approaching its lower band (12,072). These are the levels to watch — a break below would signal further downside, a bounce from these levels would confirm support.

Volume: SOXX saw volume 1.3 times its 20-day average on July 2 — elevated selling pressure. TSM saw 1.21x average volume. But NVDA volume was only 0.91x average, suggesting the selling is not concentrated in the bellwether. This is consistent with the bifurcation thesis: money is rotating out of the periphery, not fleeing the core.

Key support levels to watch next week:

• SOX: 12,000–12,100 (Bollinger lower band, prior resistance-turned-support from May)

• NVDA: $189–$190 (Bollinger lower band, psychological $190)

• MU: $940–$960 (gap fill from June 18 breakout)

• AVGO: $355–$358 (Bollinger lower band)

Catalysts: What Moves the Market Next Week

The week of July 6–10 is light on semiconductor-specific catalysts but heavy on macro data that will set the tone.

Fed minutes (Wednesday, July 8): The minutes from the June FOMC meeting are the single most important event. If they reveal greater dovishness than the meeting statement suggested — particularly any discussion of rate cuts given falling inflation and oil prices — it would be a direct tailwind for growth stock valuations. Hawkish minutes would reinforce the correction.

ISM Services PMI (early week): The services sector represents the majority of the US economy. A strong reading supports the soft-landing narrative (good for risk assets); a weak reading raises recession fears (initially negative, but could accelerate rate-cut expectations).

No major chip earnings: The next significant semiconductor earnings are TSMC (expected July 17) and ASML (July 23). This means next week's price action will be driven by macro headlines, technical levels, and sentiment rather than fundamental company news. In the absence of earnings catalysts, corrections tend to be driven by positioning flows — and with the VIX low and breadth already damaged, there is risk of further de-grossing by momentum funds.

Meta compute rental developments: Any further detail on Meta's cloud computing initiative — pricing, availability, customer demand — will be scrutinized for signals about AI compute supply-demand dynamics. If other hyperscalers (Amazon, Microsoft, Google) signal similar excess capacity, the AI demand narrative takes another hit.

China export control enforcement: Any news about license application decisions under the new Commerce Department rules (tightened May 31 to close the offshore subsidiary loophole) could move NVDA and AMD specifically.

The Week Ahead: Three Scenarios

Base Case (Most Likely): Stabilization with Bifurcation

The most likely outcome is that the SOX finds a floor in the 12,000–12,300 range early in the week and stabilizes. Falling yields, low VIX, and the July 4 weekend sentiment reset provide a constructive backdrop. However, the bifurcation intensifies: AI designers (NVDA, AMD, TSM) outperform while equipment and memory names (KLA, LRCX, AMAT, TER, SNDK) continue to underperform on capex fears.

Expected SOX range: 12,000–12,900 (roughly -5% to +2% for the week) Key levels: SOX holds 12,100 support; NVDA holds $190.

Bull Case: Oversold Bounce

Dovish Fed minutes on Wednesday, combined with technical support holding and short-covering after a two-week decline, trigger a relief rally. Stocks with the best fundamentals and most attractive valuations (MU, NVDA, AVGO) lead the bounce, while expensive names (ARM, ASML) lag.

Expected SOX range: 12,600–13,400 (roughly flat to +6% for the week) Key levels: SOX reclaims 13,000; NVDA reclaims MA20 at $203.

Bear Case: De-grossing Accelerates

Momentum funds and trend-followers, who piled into semiconductor positions during the Q2 surge, are forced to cut exposure as key technical levels break. The SOX breaks below 12,000, triggering stop losses. The Meta compute rental story gains traction, raising broader questions about AI demand sustainability. Equipment stocks lead to the downside.

Expected SOX range: 11,500–12,300 (roughly -9% to -3% for the week) Key levels: SOX breaks 12,000 support; NVDA breaks $190 and tests $180.

Positioning: How to Think About It

For active investors navigating this correction, the data points to a clear framework:

Accumulate on weakness: Nvidia and Broadcom offer the most compelling risk-reward among the AI designers. Their forward P/E ratios (15.3 and 18.6) and PEG ratios (0.60 and 0.41) are pricing in significant deceleration that AI accelerator demand doesn't yet support. Broadcom's custom silicon and networking businesses make it the most diversified AI beneficiary outside Nvidia.

High-conviction contrarian buy: Micron. At 6.5 times forward earnings with a PEG of 0.14 and revenue growth of 345%, it is the cheapest stock in the sector by a wide margin. Yes, the NAND oversupply signals that triggered the July 2 selloff are real, and near-term momentum is strongly negative (-19.6% in one week). But Micron's HBM (high-bandwidth memory) business is sold out for the foreseeable future, its DRAM pricing power is intact, and the company just posted 84.9% gross margins. The memory cycle risk applies to legacy NAND, not to the AI-critical memory that is driving Micron's results. This is a name where patience could be richly rewarded — but it requires conviction to buy into the current momentum downdraft.

Avoid or trim: ARM remains the most expensive name even after the correction, at 102 times forward earnings with a PEG of 2.6 and only 20% revenue growth. There is no margin of safety. ASML, as the sole producer of EUV lithography equipment, is structurally important but trades at 36 times forward earnings with just 13% revenue growth — its PEG of 2.65 is the worst in the group. These are the names most vulnerable to further downside if the correction extends.

Be patient with: Equipment stocks (KLA, LRCX, AMAT). The capex cycle signal is genuinely concerning, and these stocks are leading indicators. They may offer better entry points in the coming weeks as the full extent of fab spending cuts becomes clear. Do not catch this knife yet.

Hedge consideration: The JPMorgan dot-com parallel is worth hedging against, even if the macro backdrop differs. A simple hedge: long NVDA paired with short ARM would express the valuation dispersion thesis directly. Long AI designers paired with short the equipment complex (via SMH or individual equipment names) would express the bifurcation thesis.

Bottom Line

This correction is healthy, not catastrophic. After an 87% surge in six months, a 13.7% pullback is mathematically normal — it returns the SOX to levels last seen in late June. What matters is the composition.

The AI infrastructure buildout is real, the demand is verified by explosive financial results (Micron's 84.9% gross margins, Nvidia's 85% revenue growth), and the macro backdrop is supportive (falling rates, low volatility, disinflation). The correction is concentrated in the parts of the sector where the cycle is weakest (equipment, legacy memory) and in the names where valuations had become untethered from fundamentals (ARM, ASML).

Next week will likely confirm the floor. Watch the SOX 12,000 level, NVDA at $190, and the Fed minutes on Wednesday. If support holds and the Fed signals flexibility, the stage is set for the next leg of the AI trade — led by the companies actually generating the revenue, not the ones selling picks and shovels to an uncertain number of miners.