Post

SanDisk fell 8% after hours, Western Digital dropped 11% after hours. The better the performance, the harder the fall. The entire storage sector collapsed—SK Hynix dropped more than 10% intraday, Samsung fell over 6%, Kioxia plummeted more than 10%. Those holding storage stocks were completely stunned. "372% growth, isn’t that enough?" Not enough. Far from it. Where’s the problem? Two words: expectations. SanDisk’s next quarter revenue guidance midpoint is $10.55 billion, while the market expectation is $11.15 billion. A difference of less than 6%, yet the stock price crashed 9%. Western Digital is even more unfair—next quarter guidance midpoint is $4.1 billion, actually higher than analysts’ expectation of $4.06 billion. But it still fell 15% after hours. Goldman Sachs summed it up in one sentence: "The core contradiction currently facing the storage industry is not a deterioration in fundamentals, but that market expectations have excessively outpaced reality." In plain terms— It’s not that storage is failing, it’s that market expectations for storage have reached the point where "you must be perfect to the extent that even guidance has to be significantly revised upward." SanDisk has risen 460% this year, Western Digital 200%. All the positives have long been priced in. But what really alerted me was another piece of news. NVIDIA is evaluating reducing the HBM configuration of Rubin Ultra. What does that mean? NVIDIA’s next-generation AI flagship GPU originally planned to be equipped with 1TB of HBM4e 12hi memory. Now, due to extremely tight HBM supply, they are considering switching to the 8hi version or even downgrading to HBM4. Video memory capacity may be cut by one-third. The world’s strongest AI chip company is forced to "downgrade" proactively. This is not a lack of demand. This is a supply bottleneck so severe that even NVIDIA can’t handle it. This leads to a deeper contradiction— Is supply shortage a boon or a constraint? In the past two years, the market treated "supply shortage" as the biggest positive—rising prices, soaring gross margins, explosive earnings growth. But now the script has changed. Supply shortage → insufficient HBM capacity → NVIDIA forced to downgrade → AI chip performance discounted → cloud providers need to buy more GPUs → higher costs → AI infrastructure efficiency declines. Each link in this transmission chain erodes the underlying logic of the "AI storage bull market." What’s even more painful—Samsung, SK Hynix, and Micron have already sold out their HBM capacity through 2027. Customers ultimately receive only 60%-70% of what they initially applied for. Demand is still rising, but capacity is locked until 2027. This is not a "good day of supply shortage," this is a "prisoner’s dilemma strangled by capacity." The divergence is turning into a rift. Citigroup says: low inventory, supply-demand fulfillment rate dropped from 70% to 50%, prices can still rise. Morgan Stanley says: memory contract prices peak in Q4. Renqiao Asset says: the storage industry has likely already peaked. Who is right or wrong? I don’t know. But I know one thing— When everyone in an industry believes "supply shortage = perpetual price increases," the real risk is never in supply and demand itself— But in the fact that "everyone believes it." Back to investing. If you ask me: will I continue to invest in storage now? My answer is: yes, but no longer blindly buying. AI’s demand for storage is real, structural, and long-term. SanDisk has signed 10 long-term agreements covering 8 customers, with minimum contract revenue of $93.9 billion, and more than half of 2027 supply already locked in. This is not a bubble. This is real, hard demand. But stock price fluctuations never depend on "whether demand is real"—they depend on "whether expectations can get higher." When expectations are already so high that "372% growth is still not enough"— every penny you earn is a bet against the market’s most extreme optimism. $XSNDK $WDC $XSKHY #存储股财报后下挫，AI内存牛市还稳吗？