To The Moon

Storage stocks finally stopped falling.

$闪迪(SNDK)$surged approximately 26%, $美光科技(MU)$gained 18.4%, $西部数据(WDC)$ rose 15.4%, and $希捷科技(STX)$ climbed 11.4%.

Microsoft’s earnings and Amazon’s accelerating AWS growth restored confidence in the AI infrastructure trade. Oversold conditions and short covering then transformed that confidence into an explosive rally.

It was an incredible session.

But one giant green candle is not enough to confirm a reversal.

Samsung Proved the Memory Boom Is Still Real

Samsung delivered one of the strongest memory earnings reports of the cycle.

Quarterly revenue reached 171.5 trillion won, up 130% year over year. Group operating profit rose to 89.5 trillion won, while the semiconductor division alone generated 89.2 trillion won in operating profit—more than 250 times the previous year’s level.

The chip division reached a record operating margin of approximately 70%, driven by rising DRAM and NAND prices and accelerating AI-server demand.

Samsung also expects HBM4 revenue to more than triple in the third quarter as it gains ground in the AI-memory market.

These are not numbers from an industry entering a normal cyclical downturn.

The stronger signal came from Samsung’s customers.

The company has signed long-term supply agreements with the world’s five largest data-center operators and is approaching deals with another five major customers. Some contracts last at least five years and include upfront payments and minimum-price clauses.

Samsung ultimately wants roughly two-thirds of its memory production covered by long-term agreements.

Management also warned that memory shortages could become more severe and persist through 2028.

Major data-center customers are not worried about memory going unsold. They are worried about not being able to secure enough of it.

SK Hynix Explained Why Strong Earnings Were Not Enough

SK Hynix told the other half of the story. $ $SK海力士(SKHY)$

Revenue reached a record 79.3 trillion won, up 257% year over year. Operating profit climbed 557% to 60.5 trillion won, with an operating margin of 76%.

Those numbers look extraordinary.

But revenue still missed the market’s estimate of approximately 84 trillion won, while operating profit fell short of the 64 trillion won expected by investors.

The company said delays in some advanced-product shipments limited revenue recognition. Its heavier exposure to HBM also mattered because HBM prices did not rise as quickly as conventional DRAM prices during the quarter.

Net profit reached 93.9 trillion won, but a large portion came from investment gains related to the Kioxia stake rather than normal operations.

The stock fell 9.6% after the report.

SK Hynix did not fall because AI-memory demand disappeared. It fell because record results were still not good enough for an impossibly high valuation and expectations bar.

The company said major customers were continuing to request additional supply. It has completed negotiations on roughly ten long-term agreements, typically lasting five years, and plans to raise 2026 capital expenditure by approximately 50% to meet demand.

That is not the behavior of a company preparing for an order collapse.

But higher CapEx, delayed shipments, limited details about shareholder returns and long-term contracts that could reduce future pricing upside gave investors enough reasons to sell.

Samsung and SK Hynix Sent Different Market Messages

Samsung benefited more directly from the rapid price increases in conventional DRAM and NAND. SK Hynix’s heavier HBM exposure provided strong long-term growth but less immediate pricing upside.

Samsung beat an already high earnings bar.

SK Hynix produced record results but missed an even higher one.

Yet both companies confirmed the same underlying fact:

AI-memory demand remains strong, customers are requesting more supply, and major buyers are willing to commit capital years in advance.

The difference was not demand.

The difference was expectations.

Why the U.S. Storage Rally Was So Violent

Before the rebound, Micron had fallen more than 30% in July at one point. SanDisk had lost nearly half its value from its previous peak. Western Digital and Seagate had also suffered repeated selling.

The SK Hynix earnings miss accelerated the liquidation.

Leveraged traders were forced to sell. Long positions were stopped out. Quantitative funds reduced exposure. Short sellers increasingly crowded into the same trade.

Then Microsoft proved that AI infrastructure can generate revenue and cash flow. Amazon reported 37% AWS growth, its fastest pace in 18 quarters. Samsung confirmed that physical memory demand remained extremely tight.

The market suddenly realized it had pushed the “AI CapEx collapse” narrative too far.

The fundamentals provided the reason to buy. Extreme positioning turned that buying into an 11%–26% rally.

That is why the rebound was so violent—and why it is still too early to call it a confirmed reversal.

The First Pullback Will Decide Everything

The next important signal is not whether storage stocks gain another 10% immediately.

It is what happens during the first pullback.

If SanDisk, Micron, Western Digital and Seagate pull back on lighter volume, hold most of the rebound and remain above their recent lows, the rally may prove to be the beginning of a genuine bottom.

That would suggest forced selling has been exhausted and stronger buyers are beginning to absorb supply.

But if traders chase the rally and another rise in Treasury yields, oil prices or macro risk erases the entire move, then this was only an extremely powerful short-covering rally.

My View

I remain bullish on AI, and I do not believe the memory supercycle has ended.

Samsung’s record semiconductor profit proves that DRAM and NAND pricing remains exceptionally strong. SK Hynix’s results show that HBM demand continues to expand even though the company failed to clear the market’s extreme expectations.

Both companies are signing multi-year supply agreements because their largest customers want more memory—not less.

The recent collapse was mainly a valuation, expectations and positioning problem. It was not evidence that physical demand had suddenly disappeared.

But being bullish on AI does not mean calling a confirmed reversal whenever SanDisk jumps 26% in one session.

The “AI CapEx is about to collapse” thesis took a serious hit.

The storage reversal still needs to prove itself.

The answer will not be found in today’s giant green candle.

It will be found in the first pullback.