Nasdaq Tumbles 4% as Chip and Memory Stocks Sink: A $1.2 Trillion Wipeout

By Friday afternoon in New York, the arithmetic was inescapable. The Nasdaq had tumbled 4%, closing down 4.18% — its largest single-day decline since the April 2025 tariff shock — while the Philadelphia Semiconductor Index closed down 10%, its worst session since the panic of March 2020. Across the chip complex, more than $1.2 trillion in market value evaporated in a single trading day. Memory stocks, the year’s most ferocious winners, led the slide.

The rout crowned a week that had begun with Broadcom’s disappointing outlook for its custom AI chip business and ended with a far stronger-than-expected US jobs report that reignited fears of Federal Reserve rate hikes. Both legs of the trade that had carried Wall Street to record highs — AI infrastructure spending and an inevitable dovish pivot from Jerome Powell — broke at once.

For investors who had treated every chip dip since January as a buying opportunity, the assumption ran headlong into a tape that, in the words of one proprietary trader, finally refused to cooperate.

The Run That Ran Out of Road

To grasp the scale of Friday’s reversal, it helps to remember where the trade stood a week ago. The S&P 500 had just closed its ninth consecutive weekly gain — its longest streak in four decades — and the Nasdaq was within striking distance of all-time highs. The three largest memory makers on Earth, Samsung Electronics, SK Hynix, and Micron Technology, had all crossed the $1 trillion market-cap threshold, an unprecedented trio of memory giants riding a single cycle.

That convergence was not historical accident. It was the logical endpoint of a thesis that had dominated institutional portfolios for the better part of two years: that the AI build-out would generate persistent, structural demand for advanced memory — particularly high-bandwidth memory, or HBM, the stacked DRAM that sits beside every Nvidia GPU — and that supply would be unable to keep pace. Contract prices for HBM3E had risen more than 70% year-on-year, and SK Hynix alone had warned that its 2026 output was effectively sold out. For a sector long dismissed as the most cyclical corner of the semiconductor industry, the new narrative was that the cycle had, at last, been tamed.

Into that backdrop stepped Broadcom on June 3. The company’s fiscal second-quarter revenue of $15 billion met the consensus, but its outlook for AI-oriented ASIC chips, the custom silicon that hyperscalers like Google and Meta increasingly commission, failed to clear the loftiest expectations. The stock slid 14% over two sessions. The chip complex, accustomed to forgiving such stumbles, did not forgive this one. As one sell-side note circulated on Thursday morning, the issue was no longer whether Broadcom could grow into its multiple, but whether any chip company could.

On June 4, the research firm SemiAnalysis published a note arguing that Nvidia’s next-generation AI supercomputer rack, the Vera Rubin NVL72, would ship with roughly half the SOCAMM DRAM capacity that analysts had assumed — about 28 terabytes rather than 55 — because of tight memory supply. Nvidia CEO Jensen Huang denied the cuts in a June 5 interview, but the headline had already done its work. Investors, suddenly unsure whether memory would be the bottleneck everyone had assumed, began to ask what else might be mispriced.

Anatomy of the Nasdaq 4% Tumble

Friday’s session opened with the futures market flashing red and never relented. By the close, the damage was both broad and brutal. The Nasdaq’s 4.18% drop was the largest single-session decline since April 10, 2025, the day after President Trump’s “Liberation Day” tariff announcement rattled global markets. The S&P 500 fell 2.64% and the Dow Jones Industrial Average shed 1.35%, ending a winning streak that had become its own kind of market psychology.

Inside the chip complex, the dispersion told its own story. The most punished names were the most cyclically exposed. Marvell Technology fell 16%, Micron dropped 13%, and Intel, Sandisk, and Western Digital each lost roughly 11%. Qualcomm and AMD slid 10.7%. Nvidia, the bellwether of the AI trade, declined 6%. Even Oracle, more software than silicon, gave back 9.5% as investors re-priced the entire AI infrastructure stack rather than pick individual casualties. The optical communications group, often a tell for AI capex, was hit just as hard: Corning dropped 10.18%, Coherent 10.64%, Lumentum 8.62%, and Ciena 8.85%.

The Philadelphia Semiconductor Index’s 10% plunge, the worst since the Covid crash of March 2020, erased more than $1.2 trillion in market value from a sector that had added several times that amount over the preceding 12 months. For a trade that had treated memory as a structural growth story rather than a cyclical one, the cyclical was back in a single session.

The catalyst that pushed an anxious tape into a stampede came at 8:30 a.m. New York time. The Bureau of Labor Statistics reported that the US economy had added 272,000 non-farm payrolls in May, well above the 180,000 economists had forecast, while wages rose 0.4% on the month. The unemployment rate ticked up to 4.2% from 4.1%, but the headline number was unambiguously hot. Within minutes, the yield on the 10-year Treasury note climbed 11 basis points, and the CME’s FedWatch tool repriced the probability of a rate hike by year-end from roughly 22% to nearly 40%.

The bond market’s verdict was simple. If the labour market is still this tight, the Fed has no business cutting rates. And if the Fed is not cutting, the long-duration cash flows that justify a forward earnings multiple of 35 times for the Nasdaq’s biggest constituents suddenly look a great deal more expensive. The move higher in real yields — which crossed 2.1% for the first time in 18 months — acted as a gravitational pull on every high-multiple name in the index.

The Memory Reckoning

What made Friday different from a routine risk-off day was the role of memory stocks. The trio of Samsung, SK Hynix, and Micron had collectively added more than $1.5 trillion in market value over the past year, fuelled by a global DRAM and NAND shortage that pushed contract prices for high-bandwidth memory to record highs. SK Hynix shares had risen 186% year-to-date; Samsung’s memory business had helped push its parent up 114%. All three had become trillion-dollar companies on the assumption that AI capex would continue to outrun supply.

Friday suggested that assumption is no longer free.

The SemiAnalysis note, even if its most alarming projections prove wrong, introduced a doubt that the consensus had not previously entertained: that even Nvidia, with effectively unlimited customer demand, might struggle to source the memory it needs. If the AI leader is forced to cut memory configurations, the marginal DRAM bit that the market had been pricing as scarce might be less scarce than feared. The cyclical-bear case for memory — that the boom would eventually end, that inventories would build, that prices would crash — suddenly had a foothold. The fact that Huang denied the cuts on Friday morning, in a carefully worded interview, did little to allay the doubt. In markets, denials rarely travel as far as original reports.

It did not help that China’s State Administration for Market Regulation had raided the Beijing, Shanghai, and Shenzhen offices of Micron, Samsung, and SK Hynix on May 31, opening an investigation into the three companies’ control of 96% of the global DRAM market. With Seoul-listed SK Hynix down more than 8% and Samsung off 5% in Monday follow-through trading, the memory complex was repricing on three fronts at once: cyclical doubt, geopolitical risk, and AI capex anxiety. The average price of a 32GB DDR5 module — once a footnote in the AI story — has risen between 40% and 70% over the past four quarters, evidence that the supply crunch has begun to bite in adjacent consumer markets too.

The Other Side of the Trade

To be fair to the bulls — and there are still many of them — the case for owning chip and memory stocks has not collapsed. It has merely become harder to defend at any price.

Hyperscaler capital expenditure is on track to reach $750 billion globally in 2026, according to industry estimates, with more than two-thirds directed at AI infrastructure. Memory remains the single most constrained input in that build-out, and the major suppliers have publicly committed to expanding capacity — though none of the new fabs will reach volume production before late 2027. The structural shortage is real, and the demand curve is still pointed upward. Even Dennis Dick, the proprietary trader at Triple D Trading whose quoted remarks circulated on Friday, framed the moment not as the end of the trade but as the end of complacency. His comment to Caixin — that “for a long time, investors have been almost blindly buying the dip in chip stocks, and this strategy has worked. But today, that ended” — was a warning against reflexive dip-buying, not a declaration that the cycle was over.

Even Mark Hackett, chief market strategist at Nationwide, told CNBC that investors had been “hovering with their finger over this sell button” but were “not necessarily looking to get out.” Friday, in that telling, was a positioning event, not a thesis change. The economy is still growing. The AI capex cycle is still intact. The Fed, even if it delays cuts, is not raising rates into a recession.

The historical analogue is not 2000. It is the autumn of 2018, when a similar cocktail of tight labour markets, hawkish central banks, and frothy tech multiples produced a 20% drawdown in the Nasdaq that proved to be a magnificent buying opportunity. The bears were right about the correction and wrong about the cycle. They were right, too, that valuations had run ahead of fundamentals — but those fundamentals caught up within six quarters, and the index went on to triple. Memory, like any other commodity-linked corner of the technology stack, has always rewarded patient capital and punished the impatient.

A Crossroads for the AI Trade

What happens next depends on three things, in roughly this order: the next round of chip earnings, the Fed’s policy path, and whether the AI capex story reasserts itself before quarter-end.

Nvidia and Micron report in the coming weeks. If either can credibly argue that the memory bottleneck is real, that AI demand is unslackening, and that 2026 capex guidance is going higher, the drawdown will be remembered as a healthy reset. If both strike a more cautious note — particularly on memory supply — the cyclical-bear case takes over, and the next leg down could test the 200-day moving average on the SOX index, currently some 8% below Friday’s close. Option markets are already pricing the uncertainty: implied volatility on the Philadelphia Semiconductor Index rose 38% on Friday, the largest single-day jump in eighteen months.

The Fed, for its part, has limited room to reassure. With core PCE running above target and the labour market re-accelerating, the case for holding rates at their current 4.25% to 4.50% range is, if anything, stronger than it was a month ago. Powell may not need to say anything new at the June FOMC meeting; the data are saying it for him. The dot plot released in March, which projected only one cut in 2026, may now look almost aggressive.

There is also the question of liquidity. The chip sector had become extraordinarily crowded: a recent Goldman Sachs prime-brokerage report, cited by sector observers, showed net long positioning in the SOX futures complex at the 96th percentile of its post-2010 distribution. Crowded trades unwind quickly because the marginal seller has nowhere to hide. Friday’s volume on the Nasdaq exceeded 12 billion shares, roughly 70% above the 30-day average, evidence that forced selling — not merely profit-taking — played a meaningful role. The so-called “vol-mageddon” traders who had been selling call options to collect premium suddenly found themselves on the wrong side of a gamma squeeze in reverse.

For an industry that spent twelve months becoming the most beloved trade on Wall Street, the lesson of Friday is an old one. Trees do not grow to the sky. The AI build-out will continue. Memory will remain scarce. But the price of admission, after a year in which investors paid any multiple asked, has just been repriced — and the market that emerges on the other side of this shake-out will be leaner, more selective, and less inclined to mistake momentum for permanence.

And this time, the dip buyers had better do their homework.