AI Chip Stocks Shed $1.3 Trillion in Worst Semiconductor Selloff of 2026

The most significant correction in the AI investment cycle arrived without warning on June 26, 2026. Over two brutal trading sessions, semiconductor stocks shed more than $1.3 trillion in market capitalization globally — the largest two-day wipeout in the sector’s history. The Nasdaq Composite fell more than 4% in a single session. The Philadelphia Semiconductor Index, the benchmark for chip stocks, dropped 7.9%. Chip stocks as a group lost 8.1% in a single day.

For a sector that had spent the first five months of the year pricing in a future in which AI demand grows without limit, the numbers were a reckoning.

The Damage, Company by Company

NVIDIA, the defining stock of the AI infrastructure era, traded around $200 per share during the worst of the selling — placing it approximately 26% below its 52-week high of $236.26, reached in mid-May. The stock has now declined roughly 18% year-to-date in 2026, an unusual sustained underperformance for a company whose revenue is still growing at double-digit annual rates.

AMD fell 6% on the worst day before recovering most of those losses within subsequent sessions. Micron, which had been scheduled to report earnings on June 24, dropped more than 8% in pre-earnings trading — a particularly painful setup for a company that had guided strongly and was widely expected to report record quarterly revenue.

The damage extended well beyond US borders. Samsung Electronics and SK Hynix — the South Korean HBM duopoly whose memory chips power every major AI accelerator — each fell more than 12% in Seoul, dragging South Korea’s Kospi index down 10% over the two-day period. Japan’s Nikkei fell 3.55%. European chipmakers, including ASML and Infineon, lost between 5% and 8%.

The global nature of the selloff was notable. Previous AI stock corrections had tended to be US-centric. This one propagated across every geography where semiconductor manufacturing or AI infrastructure investment is a meaningful part of the economy.

What Actually Triggered It

Market participants and sell-side analysts have offered several explanations, and the evidence suggests all of them are at least partially correct.

The jobs report surprise was the proximate trigger. May employment data released on June 26 showed 172,000 new jobs added — more than double the consensus forecast of 85,000. On a normal day, a strong jobs number is bullish news. In the current macroeconomic environment, where the Federal Reserve has been carefully telegraphing a gradual rate reduction path, a significantly above-consensus employment print shifts rate expectations in a hawkish direction. Higher-for-longer rates are particularly damaging to high-multiple technology stocks, and AI infrastructure names have been trading at valuation multiples that leave no room for error.

Stretched valuations after nine consecutive weeks of S&P 500 gains provided the fuel for the correction. Before the selloff, multiple chip stocks were trading at forward price-to-earnings ratios that implied growth assumptions that even the most optimistic bull cases struggled to justify. Professional traders and institutional holders with paper gains had multiple weeks of momentum reversals in the back of their minds; the jobs report provided the excuse to book those gains.

GPU rental price deflation supplied a more fundamental concern. NVIDIA’s B200 GPU — the flagship chip running today’s large-scale AI training and inference deployments — has seen its rental price on the spot market fall from $6.11 per hour on May 30 to $4.22 per hour by June 21, a 31% decline in three weeks. Spot GPU rental prices are a real-time signal of the supply-demand balance in AI compute. When they fall sharply, it suggests that the hyperscaler build-out is adding supply faster than AI workload demand is consuming it.

This matters enormously for NVIDIA’s revenue trajectory. If data center operators are paying 31% less per hour of compute, the economic pressure to build more runs into a simple arithmetic problem: capital investment in GPU clusters becomes harder to justify when the return per unit of compute is declining.

The Bubble Question

The selloff has reignited a debate that has been simmering since early 2026: is the AI infrastructure investment cycle a bubble?

The honest answer is that the evidence cuts both ways, and the most thoughtful market participants are careful about the distinction between valuation correction and fundamental breakdown.

The bull case remains intact at the structural level. AI adoption in enterprise software, coding assistance, drug discovery, financial services, and autonomous systems is genuinely accelerating. Companies across every sector are embedding AI into their core workflows at a pace that would have seemed implausible two years ago. The demand for compute — whether measured in training runs, inference queries, or agentic task execution — is growing. The hyperscalers’ capital expenditure plans, which collectively exceed $700 billion in 2026, reflect conviction about demand that is grounded in real customer contracts and signed long-term agreements.

The bear case focuses on timing and pricing. Markets have been pricing AI infrastructure demand as if the growth trajectory from 2024 and 2025 will continue indefinitely at the same rate. That assumption is now being stress-tested by GPU spot price data, by rising competition from non-NVIDIA alternatives including AMD, Huawei’s Ascend chips, and Google’s TPUs, and by the early signs that enterprise AI adoption — while real — is taking longer to translate into the bottomline productivity gains that would justify the infrastructure investment.

Most serious analysts characterize what happened in late June as “a valuation adjustment rather than a fundamental market breakdown.” Earnings growth remains robust. Guidance stays strong. No major AI infrastructure customer has cancelled or substantially reduced its purchase commitments. But the sector has spent the first half of 2026 being rewarded for expectations; it is now being asked to produce results.

The Ripple Effects

Beyond the immediate market damage, the selloff has several practical consequences for the AI hardware ecosystem.

Secondary chip stocks suffered disproportionately. Stocks like Arm, Marvell, and Sandisk, which had benefited from the general AI-infrastructure enthusiasm, fell harder than the market leaders — Arm lost nearly 4%, Sandisk dropped 10%. These companies have less pricing power and weaker earnings visibility than NVIDIA and AMD, making them more vulnerable when sentiment turns.

South Korea’s economic exposure is now visible. The Korean Kospi’s 10% two-day drop is a reminder of how concentrated that economy’s equity market has become in AI semiconductor exposure. Samsung Electronics and SK Hynix together account for a significant fraction of the Kospi’s total market capitalization. When AI chip sentiment turns, South Korea’s entire stock market moves with it — a vulnerability that the June 28 announcement of Samsung’s $648 billion domestic investment commitment is partly designed to hedge against over the long term.

NVIDIA’s cheapness is relative, not absolute. At $200, NVIDIA is trading at roughly 25-30x forward earnings — down from the 35-40x range it commanded at its May peak, but still substantially above the long-run average for semiconductor companies. The question analysts are now asking is whether that multiple is justified by the growth trajectory, or whether the spot GPU price data suggests the growth rate is about to decelerate.

What Comes Next

The AI infrastructure investment cycle is not over. The capital commitments are signed, the data center sites are breaking ground, and the AI models requiring that compute are getting more capable every quarter. But the market has served notice that the era of rewarding AI stocks on belief alone has ended.

The next phase will be an earnings phase. The companies that can demonstrate — in quarterly results, in GPU utilization rates, in customer contract renewals — that the AI infrastructure they have built is generating the returns the investment justifies will be rewarded. Those that cannot will face sustained multiple compression, regardless of the long-term promise of the technology they are building.

For NVIDIA, the June selloff may ultimately prove to be a healthy reset. Or it may prove to be the first movement of a longer correction. Which answer is correct depends almost entirely on whether the B200 spot rental price is a temporary dip or the beginning of a structural repricing of AI compute.

That question will be answered over the next two to three quarters — and the entire semiconductor market will be watching.