Micron's earnings report eve: AI storage stocks suddenly plunge collectively

On June 23, Asian technology stocks collectively pulled back, ending an eight-day run of gains. SK Hynix fell by more than 6.5% at one point during the day, Korean stocks plunged nearly 10%, and Japanese stocks moved lower in tandem. Overnight, market sentiment shifted from excitement to caution—while everyone has been waiting for the “winning moment” is tomorrow, after the U.S. stock market closes on June 24.

Micron Technology will release its fiscal third-quarter earnings report for 2026 at that point. The market has described this report as “explosive.” Bloomberg’s consensus expects revenue of $35.02 billion, up 276% year over year; adjusted net profit of $22.96 billion, up 953% year over year; and earnings per share (EPS) of $19.98, up 1090% year over year. By comparison, Micron’s revenue for the entire 2025 fiscal year was only $37.4 billion—meaning that in just this quarter, it is essentially on track to earn roughly the amount it earned across all of last year.

If you look only at the recent picture, it can feel jarring: on the one hand, Wall Street analysts are aggressively raising their target prices; on the other hand, the stock price suddenly pulled back across the board. What exactly is happening behind the scenes?

Before going further, let’s first briefly get to know the two popular protagonists that have been in the news lately and their main businesses.

Micron Technology, founded in 1978 and headquartered in Boise, Idaho in the United States, is one of the world’s top three memory chip manufacturers. Its business spans DRAM, NAND flash, and high-bandwidth memory (HBM) used in AI chips. It is currently the only “pure-blood” storage concept stock on the U.S. stock market, and also the only major storage chip giant based in the United States.

These chip-related terms may sound unfamiliar, but they can actually be understood with a simple analogy. If AI chips are like a super brain, then the storage system is its “memory system,” but this memory is not a single structure—it is layered.

• DRAM is the brain’s “working memory” area, similar to a draft that you can keep writing and revising at any time. GPU and CPU calculations must rely on it to pull data in real time—without DRAM, even the strongest computing power can’t be fully utilized.

• NAND is the “long-term memory repository,” responsible for persistent storage of massive amounts of data, such as training datasets and model weight files, but it cannot directly participate in computation.

• HBM sits between the two, but more importantly, it isn’t just a simpler faster version of DRAM. It vertically stacks multiple layers of DRAM and directly integrates with GPUs through advanced packaging technology, forming an ultra-high-bandwidth data “transfusion channel.” It doesn’t solve a capacity problem—it tackles the most critical bottleneck in AI training: the problem of not being able to feed data fast enough to the compute units.

That’s exactly why the role of storage chips has changed over the past two years. In the past, storage was just an unremarkable component inside computers and phones. The industry overall was widely viewed as “cyclical stocks,” with prices swinging wildly up and down with supply and demand—making money depended on both luck and endurance. But after the AI era arrived, the bigger the model and the more people use it, the greater the needed storage capacity and bandwidth. Storage chips have shifted from “accessories” to a key bottleneck that determines whether AI computing power can truly come into play. This is also why Micron and SK Hynix’s stock prices and market values have become metrics that the entire tech industry is watching.

SK Hynix is the core subsidiary of South Korea’s SK Group. Its predecessor was the semiconductor division of Hyundai Electronics. After being acquired by SK Group in 2012, it was renamed to its current name. It is also one of the world’s top three memory chip manufacturers. Currently, it has the largest share in the most profitable HBM segment and is Nvidia’s most important memory supplier for AI chips.

Together, these two companies—along with South Korea’s Samsung Electronics—almost monopolize the global DRAM and HBM markets. In this round of AI infrastructure momentum, these three names in storage chips are the most central.

48 trading days — Micron completes Nvidia’s 490-day journey

To understand this pullback, you need first to see what kind of rally the market has actually been having.

On May 26, Micron’s stock price surged 18%-19% in a single day, and its market capitalization surpassed $1 trillion for the first time. This in itself was a record. From a market cap of $500 billion to $1 trillion, Micron took only 48 trading days, while Nvidia took about 490 trading days to cover the same distance, and Apple took about 1,520 trading days. Micron’s pace is ten times Nvidia’s.

The catalyst behind this surge was UBS analyst Timothy Arcuri raising his target price from $535 to $1,625 in one step, making it the highest target price among the 46 analysts covering Micron on Wall Street at the time. UBS’s reasoning was very direct: Micron has no reason not to trade at Nvidia’s valuation level. The logic is that Micron is transforming from a traditional cyclical commodity stock into a structurally growing company backed by long-term supply agreements (LTA). These agreements not only lock in procurement volumes, but also—at least partially—lock in prices, with terms that can be extended to 3-5 years. This means Micron’s historically extremely volatile earnings curve could finally be smoothed out for the first time.

Micron is not the only storage stock that has been surging wildly. Its biggest rival, SK Hynix, has also been setting records in parallel. On June 19, SK Hynix’s stock price rose 6.22%, bringing its market cap to $1.31 trillion, lifting its ranking among global assets to 17th. In the same day, Micron rose 8.7%, with a market cap of $1.278 trillion. The two companies’ market caps both exceeded Bitcoin’s market value of about $1.261 trillion at the time. Last August, Bitcoin’s market cap also once surged into the global top five. Now it has been overtaken in turn by Microsoft, Amazon, TSMC, Broadcom, and Meta—and this time, two more memory companies have been added.

Then on June 22, Micron dealt an even heavier card: it signed a long-term memory and storage supply agreement with Anthropic covering its entire product line—HBM, DRAM, and SSD. At the same time, Micron also participated in Anthropic’s just-completed $65 billion Series H funding round, with a post-money valuation of $965 billion. In that financing, the three major memory giants—Micron, Samsung, and SK Hynix—appeared on the investor list. The signal from this combination was very clear: in the AI infrastructure race, whoever can supply HBM reliably will secure a core position. On the day the news was announced, Micron’s stock rose again by nearly 7%, reaching $1,211 and setting a new all-time high. Its cumulative gain since the start of the year has already exceeded 280%.

When you connect these developments, in less than a month Micron has done three things: its market cap surpassed $1 trillion, it has deeply bound itself with top-tier AI customers, and it has pushed Wall Street’s target price to a record high. No matter how you look at it, it should be a “good news being realized” script.

But on the eve of the earnings report, the market suddenly hesitated.

Why did everyone suddenly stop daring to bet?

The optimistic side only believes one thing: supply fundamentally can’t catch up with demand.

In a podcast interview on June 5, Micron CEO Sanjay Mehrotra said there is a judgment: the AI race is not only a computing race, it is also a storage race. The larger the model, the longer the context window, and the more tokens are consumed—AI needs not only computing capability, but also “memory capability.” On the supply side, the constraint is structural. Building a wafer fab—from breaking ground to the first batch of wafer output—typically takes three to four years, and afterward it still needs to ramp up further. Meanwhile, as technology nodes advance, the incremental growth in storage capacity that each wafer can produce is getting smaller. Micron’s HBM capacity for 2026 is already fully sold out, and management expects the tight supply-demand situation for DRAM and NAND to persist beyond 2026.

Deutsche Bank raised Micron’s target price from $1,000 to $1,500 because this supply-demand imbalance will likely continue into the second half of 2026, 2027, and even 2028, and could even worsen further. TD Cowen’s view was even more aggressive: it raised the target price from $660 directly to $1,500. The core argument is that the role of memory in AI is structural rather than cyclical—meaning that this time, it really might be different.

The cautious side, however, can’t forget one thing: this industry has never escaped cycles.

Micron has been through several painful downcycles. The most recent was in 2023, when there was excess supply, causing prices to collapse and the company to record losses directly. Now Samsung and SK Hynix are both aggressively expanding HBM output. Micron’s own capital expenditures for fiscal 2026 also exceed $25 billion, and fiscal 2027 will add another more than $10 billion. In the storage industry, the worst thing—one that has repeatedly happened in history—is this: during boom periods, everyone expands capacity together, but there is always a time lag before the new capacity comes online. If Samsung, Hynix, and Micron all ramp up at the same time, and AI capex growth slows down, the pricing “leverage” that has supported high gross margins could move in the opposite direction.

There was even a small preview at the end of March. Google released a compression algorithm called TurboQuant, which can compress the KV cache that uses the most memory during large-model inference to one-sixth of the original, with almost no loss in accuracy. The market did the math: if algorithms like this are widely adopted, incremental demand for DRAM in AI would drop sharply. Within four days, prices in the consumer memory channel collapsed, falling by more than 20%. Although this shock mainly hit consumer DDR5 rather than HBM, and the market may not distinguish them strictly, it still reminded everyone of one thing: AI storage’s “structural narrative” and the idea that algorithmic efficiency could break through at any time always have a tension in them.

So this time, the real deciding factor for the earnings report comes down to one metric: gross margin. In the previous quarter, Micron’s non-GAAP gross margin jumped from about 37% a year earlier to roughly 75%. Expansions at this scale are extremely rare in a highly cyclical industry like storage. The market will focus on two things: whether management will release signals that “demand visibility for 2027 continues to improve and supply remains tight”; and whether more large customers, like Anthropic, will keep extending long-term agreements further into the future. If they simply repeat old talking points—vague about 2027 demand, pricing mechanisms, and the details of capacity ramp-up—then the “all good news is already priced in” script might truly play out on earnings night.

Bumps before dawn, or a sign of a bubble?

Tonight’s decline, based on the current picture, looks more like an early release of emotion rather than a new crack appearing in fundamentals. The fund manager Jian Shi Cortesi of GAM Investment Management explained it like this: many investors holding these AI stocks have already made a lot of money, and any small change—even a slight breeze—could prompt them to trim positions and lock in profits. Coupled with developments such as progress in U.S.-Iran negotiations, and events like SpaceX raising $75 billion despite having negative cash flow after going public—things that make the market re-examine whether “AI concept stocks” need to show real performance—investor patience is becoming fragile.

Worth noting is that this kind of tension is itself a “growth pain” that only shows up now, after the market has already run this far. Half a year ago, no one was this nervous about storage chip earnings. SK Hynix’s willingness to refuse to sell cheaply during extreme pessimism, betting on a decade of HBM until it pays off, is itself enough to show that many in this industry genuinely believe in the logic that “surviving cycles is what lets you capture the upside.” Tomorrow’s Micron earnings report, in a sense, is a test paper to verify whether this logic still holds true this time.

The “reveal” moment is after the U.S. stock market closes on June 24.