This year AI data centers absorbed nearly all available DRAM and NAND supply, more than quadrupling prices in a year:
The chain runs: AI data center demand pushes DRAM/NAND supply to its limit, quadrupling prices in a year → Micron and SanDisk gain 289% to over 2,000% this year and SK Hynix completes the largest-ever foreign IPO on Wall Street → the rally gets too fast and too concentrated, so any minor jolt (Samsung's "sell the news," SK Hynix's soft guidance) is enough to trigger profit-taking → Wall Street smells the volatility and rushes out three SK Hynix 2x leveraged single-stock ETFs (Direxion, GraniteShares, ProShares) the same week → thin ADR float plus leveraged flows turn the stock into a mechanical amplifier, multiplying every swing by more than 2x — which is exactly how you get a 27% spike followed by a 5% reversal two days apart.
The detail that matters: this isn't "the AI memory supercycle broke" — it's "a rally that ran too hot, too fast, got amplified by a batch of brand-new leveraged derivatives into something that looks like a crash." Institutional order flow from the same window backs this up: on Jul 8, Micron drew $122 million in institutional net buying (up 2.7x from the day before), while retail net-sold $202 million the same day. SK Hynix's foreign investors flipped from net-selling 542,000 shares to net-buying 88,000 on Jul 8, even as Samsung's foreign investors kept selling 3.01 million shares that same day — money was rotating within memory stocks, not exiting the sector wholesale.
SK Hynix CEO Kwak Noh-jung told Reuters: "We forecast that next year will be the worst year in the industry's history from the supply perspective," adding "we still forecast that customer demand will remain higher than our supply capacity even beyond 2030." He said this on the same day his company completed the largest-ever foreign IPO on Wall Street — not crisis messaging, but an unusually blunt admission of capacity anxiety.
Put/call ratios show exactly where the real disagreement sits: Western Digital, the most NAND-heavy of the group, carries a put/call ratio of 6.5, extreme bearish hedging, while SanDisk sits at just 0.54 and Micron at 0.71, both leaning bullish. Sentiment gauges for SanDisk (58.1) and Micron (51.3) run hot, but zoom out to the SOXX (49.1), Nasdaq 100 (43.5) and S&P 500 (36.4) and the broader market looks calm — the panic was contained to one sub-sector, not systemic. BofA raised its SanDisk price target from $2,100 to $2,500 in the same window while keeping a Buy rating, implying 41.8% upside — yet SanDisk still fell 10.62% the same day the target was raised, meaning even the bulls' own analyst wasn't fully believed by the market.
Gains: institutional buyers scooping up the dip (Micron's institutional net buying on Jul 8 ran more than 3x retail's net selling); long-term bulls betting the AI memory shortage persists past 2030; profit-takers who trimmed ahead of the "sell the news" reaction and avoided the bear-market drawdown.
Losses: retail investors who chased SK Hynix's 27% one-day spike on Jul 14 and got caught in the 5% reversal; holders of the newly-launched 2x leveraged single-stock ETFs, products with built-in decay and structural volatility that make them far riskier to hold long-term than the underlying stock; Western Digital, the most NAND-exposed name and the one the options market is pricing as the most worried, with a put/call ratio of 6.5.
Samsung, SK Hynix and Micron management all say the same thing: 2027 will be the industry's "worst year" for shortage, with demand outstripping supply capacity potentially into the 2030s.
BofA raised its SanDisk target to $2,500 mid-selloff while keeping Buy; across the entire Jul 8-15 window, not one company cut its forecast.
The panic stayed contained to the memory sub-sector — SOXX, Nasdaq and the S&P 500 barely moved, which argues against a systemic risk story.
Samsung and SK Hynix are both racing to add capacity, and new supply will eventually catch up to today's extraordinary pricing power, compressing margins from here.
The market broadly expects AI capex growth to peak in 2026 and taper afterward — if demand cools first, incoming supply expansion will land even harder.
The same day BofA hiked its target, SanDisk still fell 10.62% — if even the bulls' own bank doesn't fully trust its optimistic model, why should retail?
Overstated: coverage packaged this pullback as a uniform "memory bear market," implying the AI memory supercycle's fundamentals had broken. But across every source, not one company cut its forecast in this window — if anything, SK Hynix's CEO delivered a more pessimistic shortage warning. "Bear market" describes the price action, not the demand thesis.
Overlooked: three brand-new 2x leveraged single-stock ETFs launched right around this selloff, a structural factor that manufactures volatility on its own, yet almost no coverage treated it as a primary cause. Also overlooked: the institutional-versus-retail divergence — Micron's institutional net buying on Jul 8 ran more than 3x retail's net selling, a number that says more about who was panicking and who was calm than the headline percentage move does.
What this pullback really exposes isn't "the AI memory story stopped making sense" — it's that a sector that ran up too far, too fast, will turn any ordinary bout of profit-taking into something that looks like a crash the moment fresh leveraged derivatives get layered on top. What actually determines the next move is capacity-expansion progress and any buyback announcement in SK Hynix's July 29 Q2 report — if supply expansion genuinely starts catching up to pricing power, that's a real fundamental inflection. Until then, this reads far more like derivative-amplified emotional volatility than a collapse of the demand thesis.
When a headline says "a sector fell into a bear market," check two things first: did any company actually cut its forecast? And did the stock itself move, or a leveraged derivative riding on top of it? Also watch whether institutional and retail flows are diverging — if institutions are buying while retail is selling, emotion is probably running ahead of the fundamentals. The same discipline applies to your own positions: never hold a 2x leveraged single-stock ETF as a long-term bet on a fundamental story you believe in — the product's own structural decay will erode your thesis long before the story itself does.