Nothing about the company's business shifted that day—this was sentiment, not substance.
On July 2, 2026, Sandisk's shares fell 14% in a single session — not because the AI memory boom had broken, but because a single announcement from Meta about surplus compute capacity was enough to shake a market already trembling under the weight of its own extraordinary gains. The event is less a story about a company in trouble than about how markets, stretched by euphoria, can mistake a passing shadow for a structural collapse. Sandisk's fundamentals — nearly doubled revenue, tripling data center sales, $42 billion in contracted commitments — remained untouched, reminding us that price and value are not always the same conversation.
- A 14% single-day collapse in a stock that had already risen 700% in six months sent shockwaves through the entire memory sector, pulling Micron, Samsung, and SK Hynix down alongside it.
- The trigger was not a crack in chip demand but a Meta announcement about selling spare AI computing capacity — a piece of news that the market interpreted, perhaps too hastily, as a signal that the AI hardware shortage was easing.
- Beneath the panic, Sandisk's actual business told a different story: quarterly revenue nearly doubled, data center storage revenue tripled, and management guided for an even stronger quarter ahead.
- The $42 billion in multiyear contracted revenue represents customers locking in supply years in advance — a level of forward commitment that insulates the company from the kind of sudden demand evaporation the sell-off implied.
- The unresolved tension is cyclicality itself — NAND flash memory has a long history of booms followed by brutal oversupply crashes, and the question of whether this AI-driven cycle is different remains genuinely open.
On July 2, 2026, Sandisk's stock fell 14% in a single session, closing at $1,745. The damage spread across the memory sector — Micron, Samsung, and SK Hynix all took hits. For a stock that had already climbed more than 700% in the first half of the year, the drop was jarring enough to raise an obvious question: had the AI memory boom finally hit a wall?
The trigger came from an unexpected direction. The day before the crash, Meta announced plans to sell excess AI computing capacity to outside customers. Markets interpreted this as a sign that the acute AI hardware shortage driving the entire rally might be easing. Memory stocks absorbed the brunt of the selling, and after a run as violent as Sandisk's, profit-taking fed on itself.
But nothing about Sandisk's actual business changed that day. The company, spun out from Western Digital in early 2025, makes NAND flash memory — the storage chips that power AI data centers. In its fiscal third quarter, revenue nearly doubled to $5.95 billion, adjusted gross margins surged from 51% to 78%, and data center storage revenue tripled quarter-over-quarter. Management then guided for $7.75 to $8.25 billion in the following quarter — not the language of a company bracing for a slowdown.
More telling still is the $42 billion in multiyear supply agreements already signed — customers committing years in advance, willing to pay premiums to secure access. On a forward earnings basis, the post-plunge stock trades at roughly 12 times projected earnings, a multiple that prices in a significant downturn that has not yet arrived.
The genuine risk is the one memory investors have always faced: cyclicality. NAND flash has historically been among the most commoditized segments in semiconductors, prone to price spikes when supply is tight and brutal crashes when the industry overbuilds. The $42 billion backlog provides real insulation, but it does not rewrite that history. July 2 looks more like a valuation reset than a structural break — but the real test will come when industry capacity finally catches up with demand, as it always eventually does.
On July 2, Sandisk's stock cratered. The memory chip maker watched its shares plummet 14% in a single trading session, closing at $1,745. It wasn't an isolated incident. Across the memory sector, the damage spread: Micron fell alongside Sandisk, while Samsung and SK Hynix took sharp hits in Seoul. For a stock that had already climbed more than 700% in the first half of 2026, the drop was jarring enough to prompt a natural question from investors watching the tape: Has the artificial intelligence memory boom finally hit a wall?
The culprit, oddly, came from outside the memory industry itself. A day before the crash, news broke that Meta Platforms planned to sell excess AI computing capacity to external customers. That single announcement was enough to spook the market. If Meta could offload spare compute power, the logic went, then the acute shortage of AI hardware that had driven the entire rally might be easing. Memory stocks, which had been the hottest corner of the AI hardware trade, absorbed the brunt of the selling. What followed was a classic cascade: after a move as violent as Sandisk's year-to-date run, profit-taking feeds on itself. One bad headline becomes a stampede.
But here's the thing: nothing about Sandisk's actual business changed on July 2. The company, spun out from Western Digital in early 2025, manufactures NAND flash memory—the storage chips that increasingly power AI data centers. While the headlines obsess over high-bandwidth memory, Sandisk's product serves a different but equally essential function. AI models and their outputs need somewhere to live, and enterprise storage has become a booming business. In the fiscal third quarter ending in early April 2026, the company's revenue nearly doubled from the previous quarter to $5.95 billion. Adjusted gross margins jumped to 78.4% from 51.1% just three months prior. Adjusted earnings per share hit $23.41. The data center storage division specifically tripled, with revenue climbing 233% quarter-over-quarter to roughly $1.5 billion.
What matters more than any single quarter, though, is what's already locked in. Sandisk has signed multiyear supply agreements worth approximately $42 billion in guaranteed minimum revenue. These are customers committing years in advance, willing to pay premiums to secure supply. Management then guided for the fiscal fourth quarter at $7.75 billion to $8.25 billion in revenue—well above what the company had just reported. That guidance doesn't sound like a company bracing for a demand cliff. It sounds like a company running at full throttle.
The real question, then, may not be whether the AI boom is cracking but whether certain stocks riding that boom have simply become overvalued. After the plunge, Sandisk trades at roughly 59 times trailing earnings. That multiple looks expensive until you consider the trajectory. Management's guidance for the just-concluded fiscal fourth quarter alone calls for as much as $33 in adjusted earnings per share. Annualize that pace, and the stock is trading at little more than 12 times forward earnings—a multiple that prices in a significant downturn that hasn't yet materialized.
Memory chips, though, have always been a cyclical business. NAND flash in particular has historically been one of the most commoditized, price-sensitive segments in semiconductors. Prices spike when supply is constrained, then crater when the industry overbuilds. The $42 billion backlog provides real insulation, but it doesn't eliminate the fundamental risk of a sector that has punished investors before. The bet embedded in Sandisk's forward valuation is that this AI-driven cycle runs longer and more steadily than previous ones.
So was July 2 the moment the boom cracked? Probably not. The sell-off looks more like a valuation reset during a shortage that remains very much intact. The demand signals are still strong, margins are still surging, and that $42 billion backlog still exists. But calling the dip a clear bargain would be premature. Memory stocks have a long history of teaching investors that the top of a cycle can feel a lot like a new normal. The real test will come when the industry's capacity catches up with demand—and history suggests it always does eventually.
Notable Quotes
The demand signals, the surging margins, and that $42 billion backlog all point in the same direction—the boom is not cracking.— Analysis of Sandisk's fundamentals
The Hearth Conversation Another angle on the story
So Meta announces it's selling spare compute capacity, and suddenly Sandisk loses 14% in a day. That seems like an overreaction to a single piece of news.
It is, in a way. But remember how far the stock had already run—700% in six months. After a move like that, the market is primed for any excuse to take profits. One headline becomes a stampede.
But the company's fundamentals didn't change. Revenue nearly doubled, margins jumped, and they have $42 billion in locked-in contracts. How does that square with a 14% drop?
It doesn't, really. That's the point. This wasn't a crack in the boom. It was a sentiment reset. The market repriced the stock after a red-hot run, not because demand softened.
Then why not just buy the dip? If the fundamentals are that strong, isn't this a gift?
Not necessarily. Memory chips are cyclical. Prices spike when supply is tight, then crater when the industry overbuilds. That $42 billion backlog helps, but it doesn't eliminate the risk. The stock is pricing in a very long, very steady AI cycle. History suggests those always end.
So you're saying the boom isn't cracking, but the valuation might be?
Exactly. The question isn't whether demand is strong now—it clearly is. The question is when the cycle turns. And that's always been the hard part with memory stocks.