AI Chip Stocks Enter Bear Market: $1.5 Trillion Wiped in the Industry's Sharpest Correction

When markets closed on June 25, the VanEck Semiconductor ETF (SMH) had just posted a 71% gain for the second quarter — the largest quarterly return for the index in its history. Two weeks later, that same index had given back a significant portion of those gains. The AI chip trade, which had made semiconductor stocks the defining equity story of 2025 and early 2026, suddenly looked overextended.

The selloff that swept through global markets in early July 2026 erased more than $1.5 trillion in combined market value from semiconductor companies. It was the sharpest correction the sector had seen since the early months of 2023, and it arrived at a moment of maximum complacency.

How It Started: The HBM Slowdown Report

The selloff’s ignition point was a report, circulating among institutional investors in late June, that SK Hynix — the South Korean memory giant and the market’s single most important supplier of high-bandwidth memory for AI servers — was considering a moderation of its HBM production ramp. SK Hynix had been operating at full capacity to supply Nvidia with HBM stacks for the Blackwell and Vera Rubin GPU families. The suggestion that it was preparing to slow that expansion implied one of two things: either demand was softening, or yield and supply concerns were forcing a temporary pullback.

Either interpretation was bad for chip stocks. If demand was softening, the AI infrastructure build cycle that had driven semiconductor revenue to record levels was further along than expected. If it was a supply issue, margins would be pressured as the company navigated technical constraints.

SK Hynix and Samsung both fell more than 9% and 7% respectively at the open of Asian trading on July 2, dragging down South Korea’s Kospi Composite by 10% at intraday lows — severe enough to trigger automatic circuit breakers that halted trading. Japan’s Nikkei 225 fell 3.6%. Hong Kong’s Hang Seng dropped 1.8%. The cascade reached U.S. markets by the open, with the Nasdaq Composite falling 2.2% and the S&P 500 shedding 1.5%.

The Anatomy of the Decline

The single-session numbers were dramatic. Micron Technology, which had surged 884% over the prior year to a market capitalization exceeding $1 trillion — a milestone it crossed in May 2026 — fell 13% in one day, erasing approximately $138 billion in value. AMD dropped 7%. Intel fell 9%. The SMH ETF, that measure of the sector’s overall health, lost 5% before recovering some ground by close.

The second wave hit on July 7, when Samsung reported quarterly earnings that fell short of elevated expectations. Revenue and operating profit were strong by historical standards, but the market had priced in perfection. Samsung shares fell another 7% on the disappointing guidance, and the ripple touched SK Hynix, Micron, and TSMC, each of which declined between 3% and 6% in sympathy.

By July 8, Micron, Samsung, SK Hynix, and the Roundhill Memory ETF were all down more than 20% from their recent closing highs — the technical definition of a bear market. The total market value destruction across semiconductor equities since June 25 exceeded $1.5 trillion, with Micron alone accounting for nearly $350 billion of that decline from its peak.

Three Catalysts Converging

The magnitude of the selloff reflected not one problem but three converging pressures that arrived simultaneously.

The Fed tightening signal. Federal Reserve Chairman Kevin Warsh, who replaced Jerome Powell earlier in 2026, has consistently signaled a more hawkish disposition toward monetary policy. In late June, Fed minutes revealed that nine of eighteen policymakers now supported rate hikes in 2026 — up from zero supporters in March. The 10-year Treasury yield climbed to approximately 4.48%, and risk assets across equities felt the rotation. Technology and semiconductors, with high valuations predicated on future earnings, are particularly sensitive to discount rate moves.

The ROI skepticism loop. The question that institutional investors had largely suppressed throughout the AI bull run — “when does the return show up?” — resurfaced in July with unusual intensity. Microsoft, Amazon, Alphabet, and Meta have collectively committed hundreds of billions of dollars to AI infrastructure. Microsoft disclosed alone that it spent $28 billion in just one quarter on capital expenditures, almost entirely for AI. The revenue from AI products, while growing, has not yet scaled proportionally to these investments. Analysts at several major banks published notes in early July questioning whether the AI revenue models could justify current infrastructure spending levels — and by extension, current chip valuations.

The Meta neocloud signal. A story that landed in the same week as the selloff — reports that Meta was exploring selling excess AI compute capacity to outside customers — was interpreted by some investors as a sign that hyperscalers were approaching saturation in their GPU procurement. If Meta had spare capacity to sell, its orders to chip suppliers might slow. The inference was debatable, but in a market already primed for negative news, it accelerated the selling.

The Counter-Narrative: Micron’s $250 Billion Bet

Not everyone interpreted the signals the same way. On July 9, the same day the selloff was still fresh in institutional memory, Micron poured the first concrete at its new New York fabrication facility and announced an increase in its U.S. investment commitment to over $250 billion through 2035. Micron’s stock responded by rising approximately 7% — a sharp contrast to the broader sector.

The divergence is instructive. Micron’s fiscal Q3 2026 revenue of $41.46 billion, a 345.7% year-over-year increase, demonstrated that the AI memory demand cycle is currently producing real earnings, not just promise. The company’s CEO, Sanjay Mehrotra, used the groundbreaking as an opportunity to restate the long-cycle thesis: AI infrastructure build-outs, particularly in the United States, will require enormous quantities of high-bandwidth memory for years, and Micron intends to supply a significant share of it from domestic fabs.

This is the fundamental disagreement the July selloff revealed: whether current conditions represent a peak in the AI chip cycle or merely a correction within a longer structural buildout. The bears see an industry that got far ahead of itself on AI revenue expectations and must now reckon with a slower path to returns. The bulls see the same industry experiencing the growing pains of a technology transition that is still, by most measures, in early stages.

What Semiconductor Stocks Need to Recover

For the sector to stabilize and resume its upward trajectory, several conditions probably need to be met.

First, the HBM slowdown story needs clarification. If SK Hynix provides guidance that confirms sustained demand through 2027, the catalyst for the selloff weakens. If Samsung’s next earnings cycle shows strong forward orders, the earnings disappointment of July 7 becomes a one-quarter event rather than a trend.

Second, at least one major hyperscaler needs to demonstrate meaningful AI revenue acceleration. Microsoft’s Copilot products, Amazon’s Bedrock platform, and Google’s AI Overviews are all growing — but the market wants to see the pace of growth justify the infrastructure spend.

Third, the Fed’s rate trajectory needs to stabilize. A clear signal that the next move is a cut rather than a hike would relieve pressure on technology valuations broadly and semiconductor stocks specifically.

The sector’s fundamentals — unit demand for AI chips, memory prices, capacity utilization at leading fabs — remain strong by historical standards. What broke in July was sentiment, and sentiment is easier to restore than fundamentals are. But the speed and scale of the correction serves as a reminder that even the most compelling structural technology stories can generate violent price corrections when expectations get too far ahead of the underlying business.

AI chips are still the most important trade in global equities. The question July raised is not whether that is true, but whether the market had been pricing them as if that were true forever.