Author: Block Analytics Ltd X Merkle 3s Capital

Let’s start with a counterintuitive picture.

The moment the earnings report came out, the numbers were stunning—record revenue, gross margins at multi-year highs, and earnings per share far exceeding even the most bullish Wall Street estimates. Social media erupted with cheers, retail investors started screenshotting their positions, and headline writers had already teed up the phrase “money-printing machine.”

And then? The first reaction from smart money was to reduce positions.

This isn’t speculation—it’s exactly what happened last quarter. Last quarter, Micron delivered a blockbuster earnings report: revenue above guidance, profits above expectations. Yet on earnings day, the stock didn’t rise; it fell. Within about a week, it had pulled back nearly 30%. A historic report card was met with a historic sell-off.

Why? Because the market never pays for “how much you earned in the past,” but for “whether you can keep earning this much in the future.” For a company branded a “cyclical stock,” the best earnings often arrive just before the worst turning point. The moment profits peak is also the most dangerous moment for valuation logic—this fear is etched into the semiconductor industry’s bones.

So this time, we’re not rushing to applaud.

We want to ask an even more critical question: Are the massive profits in this memory cycle yet another dying gasp of a cycle, or a genuine paradigm shift? Is this the same old story of a rising tide that will eventually recede, or has the sea level been permanently raised?

The market is still using the old “cyclical” yardstick to price Micron. But if the yardstick itself is wrong, cheap could be a trap, and expensive could be the starting point.

In this article, we’ve prepared three yardsticks to test, one by one. First, we’ll lay out the earnings numbers, then dissect the valuation, and finally examine the variable that isn’t in the earnings report but could determine everything.

The Numbers Unveiled: Laying This Earnings Report on the Table

Before passing judgment, we’ll do the simplest thing—honestly lay out the numbers. No emotion, no conclusions, just let the facts stand firm first.

This quarter (FY2026 Q3), Micron’s core figures are: revenue $41.456 billion, up 346% year-over-year and 74% sequentially; gross margin 84.9%; EPS $25.11; free cash flow $18.3 billion; net profit $28.86 billion.

That one line already says enough—this isn’t a “pretty good” earnings report; it’s one that redefines Micron in magnitude.

To understand the weight of these numbers, you have to look at what they’re being compared to. A report “beats expectations” not by comparing to last year, but by measuring against three anchors: the company’s own prior guidance, Wall Street’s market forecasts, and the sequential performance from the previous quarter. We’ve laid out these three dimensions in a table.

The line in the table truly worth staring at is gross margin.

Record revenue is nothing unusual for a memory manufacturer in an upcycle—volumes rise, prices rise, and revenue naturally looks great. But gross margin is another story. Gross margin measures: “For every $100 of chips sold, how much is left as pure profit?” It directly reflects pricing power. Whether a company can keep the profits from price increases in its own pocket, rather than being squeezed by upstream equipment suppliers and downstream customers, depends entirely on this number.

Historically, Micron’s gross margin has been infamous for its roller-coaster ride. At the industry’s worst, it was selling chips at a loss, with gross margins in the single digits or even negative; at the best of times, it rarely managed to stay high consistently. So if this quarter’s 84.9% holds, its significance goes far beyond “they made a lot of money”—it’s answering a deeper question: Has Micron truly grasped pricing power this time?

Besides the main table, there are a few more numbers we suggest you note separately.

• Progress in HBM business: As the core of this AI narrative, HBM’s revenue share, shipment cadence, and HBM4 mass-production timeline are key to judging how long the “AI bonus” can last. This quarter, HBM4 12-high’s production ramp is already twice as fast as HBM3E 12-high’s, and it is expected to reach mature yield rates significantly quicker than HBM3E. HBM4 revenue already exceeded $1 billion this quarter.
• Next quarter’s guidance: Next quarter (FQ4-26) revenue guidance of $50 billion ± $1 billion, gross margin approximately 86%, EPS $31.00 ± $1.00, capex approximately $10 billion. The market always lives in the future—this quarter’s greatness is nothing compared to one hint from management about next quarter.
• Capex arrangements: FQ3 net capex $7.1 billion, FQ4 expected around $10 billion, full-year FY2026 approximately $27 billion. We’ll address this number later—it might be the most overlooked yet deadliest line in this report.

The numbers are laid out. Objectively, this is a report card worthy of Micron’s history books.

But we must give you a dose of sobriety ahead of time: The past eight quarters have repeatedly proven that what dictates Micron’s stock price has never been the earnings themselves. Whether revenue sets records or beats expectations and whether the stock rises after earnings are almost two completely unrelated things. We’ll leave this for the “valuation trap” chapter, where we’ll dissect it using eight real earnings reports.

The prettier the report card, the more cautious we must be—an epic earnings report never automatically translates into epic returns. What stands between is which yardstick the market uses to price it.

Next, let’s take apart that yardstick.

The Market Is Using the Wrong Yardstick

Open any market app, and you’ll see Micron’s forward P/E looking enticingly low—based on future earnings, it’s probably only a little over 10 times.

For most stocks, a 10x P/E is synonymous with “cheap.” A tech company whose performance is still exploding, yet whose valuation is only in the single digits—that sounds like a pie falling from the sky. Many get drawn in this way.

But for cyclical stocks, that’s precisely the most dangerous signal.

We need to be clear about something counterintuitive first: A low P/E for a cyclical stock isn’t cheapness; it’s a warning that earnings have peaked.

The logic goes like this: P/E equals stock price divided by earnings per share. When a cyclical company is at the peak of its cycle, the denominator—earnings—is pushed to an all-time high. So even if the stock price isn’t cheap, the calculated P/E will be very small. The market isn’t stupid; it compresses the valuation multiple precisely because it knows that these high earnings can’t be sustained, and they’ll likely fall next year or the year after. A low P/E is the discount the market applies in advance for “profits are about to decline.”

Conversely, at the bottom of a cycle, the company might be losing money or barely profitable, the denominator approaches zero or even negative, and the calculated P/E will be absurdly high, even exceeding 100x. At that point, the stock looks “so expensive it’s unlovable,” but that’s often the golden opportunity.

Don’t believe it? Let’s look at Micron’s own history.

In 2018, at the top of the memory cycle, Micron’s earnings were exploding. Its forward P/E was compressed to less than 3 times—nearly the “cheapest” large-cap stock in the entire market. What happened next? Over the next year and a half, the stock price fell about 70% from its peak. Those who bought at that “cheap” 3x P/E were shell-shocked.

Around 2024, the industry was just emerging from its deepest trough, earnings hadn’t yet recovered, and the P/E was calculated at over 100 times—looking absurdly expensive. What happened then? After that point, the stock mounted a nearly tenfold rally.

In the world of cyclical stocks, the rule is cold: Buy at high P/E, sell at low P/E.

That’s the exact opposite of what every investing textbook teaches you—“buy cheap, sell dear.” Because textbooks talk about companies with stable earnings, while a cyclical’s earnings themselves breathe. What you think is “expensive” is it gasping in the trough; what you think is “cheap” is it holding its last breath at the peak.

So when someone today points to Micron’s forward P/E of a little over 10 and says “severely undervalued, buy with eyes closed,” our first reaction isn’t excitement but wariness. If Micron is still that old cyclical, then this low P/E is a rerun of 2018—the market pricing in an earnings collapse in advance.

But—and this is the crucial twist of the entire article—what if Micron is no longer that old cyclical?

What if this round of high profits isn’t a flash in the pan at the cycle peak, but is structurally anchored by long-term agreements, AI demand, and the supply landscape? Then this 10x P/E is no longer an alarm but a genuinely cheap stock mistakenly killed by the old framework.

One number, two fates. Whether 10x P/E is sweet or poison doesn’t depend on the number itself, but on what kind of company you judge Micron to be.

To make that judgment, looking at P/E alone is useless. We need to take out three sharper yardsticks to measure its bones—the duration of orders, the resilience of profits, and the supply landscape.

First yardstick: measure the orders.

First Yardstick: Can Locked-In Orders Lock Down the Cycle?

Why has the memory industry had the word “cyclical” etched on its face?

Because it used to be a “spot business.” Chips were like commodities—priced as they were that day, customers bought as needed, manufacturers sold as produced. When demand heated up, everyone rushed to expand capacity; when capacity came online and oversupply hit, prices collapsed; when prices crashed, manufacturers cut production and slashed investment; after supply contracted, prices rose again—an endless cycle, with no one able to see beyond six months. This “can’t see the future” business model was the root of the cyclicality.

But this round, something is quietly changing the rules of the game—Long-Term Agreements (LTAs).

In one sentence: An LTA is when a customer no longer “buys as needed” but signs a long contract with the manufacturer a year or even years in advance, locking in quantity, locking in price, and sometimes paying a prepayment upfront. In plain terms, it’s turning a “spot stall” into an “annual canteen subscription.”

However, what’s even more worthy of our attention is something that goes a step beyond an ordinary LTA—Strategic Customer Agreements (SCAs).

The difference between it and an ordinary LTA is not just a more intimidating name. Traditional LTAs are often short-term, relatively loose “intent-based volume locks”; whereas an SCA is a deep, multi-year binding commitment — some are signed for as long as 5 years. Pushing the business model from a “yearly meal plan” one step further, an SCA is more like a client and a vendor entering a “five-year marriage pact” together — it’s not just about buying more this year, but tying the long road ahead together.

But the phrase “multi-year” carries vastly different weight. A truly ironclad SCA and a vague “we’re optimistic about long-term demand” are separated by the details hidden in contract clauses. So for this earnings report, whenever management talks about long-term agreements, we need to scrutinize it like lawyers reviewing a contract, word by word, for the following items:

• Is there a take-or-pay clause? This is the most critical one. Take-or-pay literally means “either take delivery or pay anyway”—even if the customer doesn’t take all the shipments, they still have to pay. With this clause, a long-term agreement transforms from “I plan to buy” to “I must pay even if I don’t buy,” and the manufacturer’s revenue truly gains a legal backstop.
• Is there capacity reservation? Has the customer locked in a specific production line or a certain percentage of Micron’s capacity exclusively for themselves? Once capacity is reserved, that supply is effectively withdrawn from the market in advance.
• Are there prepayments? Is the customer willing to put up real money upfront to secure a place? The size of prepayments is the most honest thermometer of demand strength—talk is cheap, upfront payment counts.
• Is there a price floor? Even if spot prices fall in the future, does the contract have an unbreakable minimum price? With a price floor, gross margins won’t be wiped out by a single demand fluctuation.
• Are there cancellation penalties? What price would a customer pay if they tried to renege or cut orders midstream? The heavier the penalties, the more genuine the order duration.

Together, these five questions essentially ask one thing: Is this long-term agreement a gentlemen’s handshake or a contract with teeth?

Why are we so fixated? Because it directly determines how much of a valuation multiple the market is willing to give Micron.

The reasoning is plain. If management can convince the market that a significant proportion of 2027 profits is “already locked in via signed contracts”—with take-or-pay backing, a price floor underneath, and cancellation penalties—then the certainty of those profits is no different from a software company’s subscription revenue. The market always pays far more for certainty than for cyclicality. Once the concept of “contracted profits” takes hold, the old 10x P/E yardstick completely breaks down.

Conversely, if management just keeps repeating “demand is very strong, customers are very enthusiastic” without ever disclosing a single concrete clause, then we need to dial down our excitement—vague optimism cannot support a revaluation thesis.

What can support a higher valuation is never a mere “strong demand” but those few clauses with teeth in the contract: take-or-pay.

The destructive power this has on cyclicality is far greater than it appears.

First, it allows manufacturers to “see the future” for the first time. Before, Micron could plan only for the next quarter at best; now, for high-end products like HBM, order visibility has been extended far out—Micron’s 2026 HBM capacity is reportedly close to 100% sold out, with some orders even scheduled into 2027. This means before the factory even runs at full throttle, the work for the next one to two years is already booked.

Second, it flips the bargaining dynamic. In the spot era, manufacturers begged customers to buy; in the SCA era, customers fear not being able to buy and are willing to lock in supply with prepayments. The balance of power has shifted from buyer to seller.

This change isn’t unique to Micron. We see industry-wide signals: SK Hynix is rumored to have signed multi-year DDR5 supply agreements with mega-customers like Microsoft, and Samsung is also using long-term contracts to lock in large clients. When memory factories’ capacity gets carved up in advance by multi-year agreements, the classic script of “oversupply leads to collapse” is being rewritten, at least in the premium segments.

And long-term agreements bind far more than price and quantity. Here’s a particularly intriguing detail: Micron is collaborating with Anthropic to specifically optimize the memory subsystem for Claude’s training and inference—tuning HBM bandwidth, suppressing power consumption, and improving overall efficiency, all tailored to the actual needs of this top-tier AI company.

What does this reveal? It shows the relationship between top-tier clients and Micron is no longer a buyer-seller relationship of “I place an order, you ship.” It’s a co-development relationship where “we customize the memory together for my model.” Once a memory subsystem is deeply optimized for your model, the cost and risk of switching suppliers becomes absurdly high.

What the long-term contract truly locks in is never just price—when memory starts being tailored for a particular company’s models, what’s locked is technology, the high wall of migration, and a deep bond that makes walking away impossible.

This brings us to the core question the first yardstick must measure—can locked-in orders really lock down the cycle?

We lean toward splitting the answer into two halves, rather than black or white.

The half that can be locked in is the high end. High-bandwidth memory like HBM, used for AI training, has high technical barriers, scarce capacity, and a highly concentrated customer base (just a few AI chip giants). These products naturally suit long-term contracts—customers fear supply disruption, manufacturers need to recoup massive R&D and capacity investments, and both sides have every incentive to make the relationship long term. In this segment, pricing power is indeed undergoing a structural shift.

The half that cannot be locked in is standard products. Ordinary DDR5 and memory used in phones and PCs remain relatively standardized commodities. They can be partially locked via long-term contracts, but can never fully escape the gravity of spot prices. If consumer electronics demand weakens, the cyclicality in this part will return.

So SCAs don’t eliminate cycles; they “layer” the cycle—the high end is propped up by long-term contracts, while standard products still swim in the water.

That’s why we must read every word management says about long-term agreements in this earnings report. And this time, management provided details that exceeded expectations: Micron has signed 16 Strategic Customer Agreements (SCAs), covering data centers, consumer electronics, and automotive, typically for five-year terms (2026-2030), with automotive around three years. These agreements together cover roughly 20% of DRAM output and about one-third of NAND output; among them, the cumulative minimum price revenue of 14 SCAs totals around $100 billion, with customer prepayments and commitment fees reaching a staggering $22 billion. Even more crucial is the structure—these agreements adopt take-or-pay, and the largest one sets both a price ceiling (locked at the current CQ2 market price) and a price floor. Management expects that once all SCAs are in place, approximately half or more of the company’s revenue will be covered by SCAs.

The points we were going to scrutinize word by word—what proportion of capacity is covered by long-term agreements? Are they locking quantity or also price? Do they have take-or-pay, price floors, and cancellation penalties? How clear is the visibility into 2027? Has prepayment size increased?—this time, almost every one got a positive answer:

Capacity is solidly locked away at one-fifth to one-third, prices are held by floors and ceilings, take-or-pay provides the safety net, $22 billion in prepayments are in the bank, and visibility stretches directly to 2030. When “a significant proportion of profits are already contracted” turns from a slogan into $100 billion in contract value, that’s one of the hardest pieces of evidence for a paradigm shift, not vague “strong demand.”

The order duration yardstick measures whether “the future can be seen.”

But seeing orders doesn’t guarantee profit. The second yardstick will measure something else—the “quality” of this round’s windfall profits.

Second Yardstick: Super Cycle or a Repeat of the Old Cycle?

The most fiercely debated term in the market right now is “super cycle.”

It implies that the memory demand driven by AI this time is not an ordinary cyclical upswing, but a “super” upswing of greater magnitude, longer duration, and one that can break old patterns. Sounds beautiful. But we first need to establish a judging criterion—otherwise, the debate is just people talking past each other.

Our criterion is simple yet harsh: To distinguish a super cycle from a recycled cycle, don’t look at how much prices rose; look at whether gross margin falls.

Why gross margin and not price?

Because prices rise in every ordinary cycle—that’s not unusual. What is truly unusual and can differentiate a “paradigm” from a “last gasp” is whether manufacturers can hold onto those high profits when demand ebbs. If this round is truly structural, then even if demand fluctuates in a given quarter, gross margin should show resilience, not revert to square one overnight. Conversely, if gross margin crumbles at the first sign of trouble, then at its core it’s still the same old cycle, just with a higher wave this time.

So we’re fixated on one number—the 84.9% gross margin, and its trajectory over the next few quarters.

First, let’s look at the evidence supporting the “super cycle” camp—and it’s far from weak.

• Data centers are swallowing most of the memory manufacturers’ capacity. AI servers have a voracious appetite for memory, many times that of traditional servers. Industry-wide, data-center-related demand consumes a very large proportion (around 70% level) of DRAM capacity. This quarter, Micron’s data-center revenue alone exceeded $25 billion, annualizing above $100 billion—itself a stunning growth curve.
• AI server shipment growth is staggering, more than doubling year-over-year (approaching +180% level), and each AI server is a memory “vortex.”
• Although HBM’s revenue share isn’t the biggest, it’s extremely “wafer-hungry”—the effective capacity yielded from making HBM on the same wafer is far lower than from ordinary DRAM. Industry-wide, HBM accounts for about two-tenths (~23%) of wafer capacity. This effectively withdraws supply from standard DRAM, supporting overall prices.
• There’s one more intuitive thermometer—inventory turnover. In a healthy state, memory manufacturers’ inventories are about 8 to 12 weeks of shipments; now, some high-end product inventory has been compressed to 3 to 5 weeks. Goods are snatched up the moment they come out—the most honest signal of a supply-demand imbalance.

Put these pieces together, and the “super cycle” story is indeed coherent: AI has created a rigid, long-term memory demand that didn’t exist before, structurally lifting the entire industry’s profit baseline.

But here, we must dispel one of the most widespread misconceptions — many people believe that the enormous profits this cycle are entirely thanks to HBM, but that’s actually not the case.

A counterintuitive fact, yet one that management has implicitly disclosed: right now, general-purpose DRAM—plain vanilla DDR5—delivers higher profit margins than HBM. In other words, the HBM that every report has been touting isn’t even the most profitable segment when you look purely at earning efficiency.

What does that imply? It means a huge chunk of that dazzlingly high gross margin isn’t coming from HBM at all, but from the extreme scarcity of traditional DRAM. As mentioned earlier, the Big Three have been shifting wafers to HBM, squeezing the supply of ordinary memory ever tighter – when a commodity becomes scarce, its price rises, so the price and profit margin of standard products get pushed to historic highs.

And that is precisely the most underappreciated piece of evidence for a “super cycle”: this round’s payoff is not a one-product show driven solely by HBM, but an industry-wide shortage across the entire memory stack. From high-end HBM all the way down to the most basic DDR5, just about every shelf is raising prices and making money. A boom supported by a single star product is fragile; a boom where every category benefits simultaneously rests on far more solid ground.

So where is HBM’s real significance? The answer lies in a deeper technical logic – Token Economics. And this is the hardest piece of technical evidence behind our view that “this time, demand might genuinely be different.”

Let’s unpack it in plain language. In the age of Agentic AI, AI no longer just answers a single question; it has to reason continuously, call tools, and run many steps on its own. The core metric for how good this AI is becomes one thing: token throughput – essentially, the speed at which the AI “spits out words,” how much content it can process and generate per second. How fast an AI can think and how many users it can serve per second all hinge on this.

And what determines token speed is an equation so rudimentary it’s almost brutal:

Token speed = HBM capacity × HBM bandwidth.

Translated into plain terms: for an AI model to think, it first has to move massive parameters and context from memory into the compute chip. HBM capacity determines how large a model and how long a conversation can be loaded in one go; HBM bandwidth determines how fast that data gets moved. These two factors directly impose the ceiling on how fast an AI can “spit out tokens.”

Many will ask: can’t software optimization do the job? Aren’t algorithms getting smarter so they can use memory more frugally? They can, but there are limits. Software optimization can maneuver beneath that ceiling, but it can never break through a ceiling that is ultimately set by hardware. As long as AI wants to run faster and serve more people, the only option is to stack more, and faster, HBM. This is not a choice of business model; it’s a constraint imposed by the laws of physics.

This is the hardest justification for why HBM demand “might genuinely be different”: it isn’t inflating because of a short-term capex impulse; it’s firmly anchored by the foundational formula governing AI performance. For as long as AI needs to get stronger, the thirst for HBM won’t stop.

There’s another piece of the puzzle that gets equally ignored – data center SSDs and NAND.

All eyes are on HBM, but few notice that AI is quietly igniting demand for enterprise flash. AI inference requires pulling up vast amounts of data; vector databases need to store an astronomical number of vectors; and there’s so-called KV cache offloading (simply put, taking parts of an AI’s “short-term memory” that are temporarily not needed and moving them from expensive memory onto SSDs for temporary storage) – all of these are consuming enterprise SSDs at a frantic pace. This quarter, Micron’s data center SSD revenue has already surpassed $5 billion, doubling quarter-on-quarter.

In other words, it has never been only HBM that enjoys the AI dividend. From HBM, to general-purpose DRAM, to data center SSD/NAND – the entire memory stack is riding a rising tide together. This further solidifies our judgment: this round is a structural boom for the whole industry, not a flash-in-the-pan for a single product.

But when we invest, we can’t just listen to one side of the story. The bearish camp’s logic is equally sober and sharp.

Their core thesis boils down to one sentence: high gross margins are themselves the seeds of a cycle reversal.

The logic is almost painfully simple – when Micron, SK Hynix, and Samsung all realize that HBM and high-end DRAM are generating obscenely fat profits, what are they going to do? Expand production like crazy, of course. And semiconductor expansion has a deadly time lag: the decision to build a fab and order equipment is made today, but the actual capacity only comes online a year or two later. When this wave of expansion lands in a concentrated way around 2028–2029, supply will suddenly flood out, and prices and gross margins could turn downwards – with multiple new fabs hitting volume production exactly in that window, that’s when the risk of overcapacity peaks. This is the standard script of every past cycle collapse – not because demand vanishes, but because supply overshoots.

Thus, some investors extrapolate from here the well-worn path of “peak margins → expansion → overcapacity → price retreat.” This logic has a clear endpoint: today’s high gross margins are unsustainable, and the market will sooner or later reprice this as a “cycle peak.”

But what’s intriguing is that this bearish logic itself is starting to crack. Even the most conservative voices that once insisted on pricing Micron at “cycle peak” have recently begun to change their tune en masse – their latest models no longer bet on “oversupply next year,” instead expecting the DRAM supply-demand tightness to persist all the way until 2028. When even the staunchest bears are revising up the supply-demand gap and repeatedly pushing back the turning point, that fact in itself is one of the most powerful pieces of corroborating evidence for the “paradigm” camp.

At its core, the divergence between the two camps is two different narratives spun around the same set of facts. We’ve laid it out in a comparison table for you to weigh for yourself.

So what’s our own view?

We’re not taking sides, but we offer a judging framework: this round is most likely a “half super-cycle” – the high end is a paradigm shift, while standard products are still cyclical.

The HBM piece, backed by technical barriers, long-term agreements, and a concentrated base of AI customers, and – even more critically – propped up by that physical formula for token speed, looks closer to a structural paradigm shift, and its gross margins will have genuine resilience. The standard DRAM piece, on the other hand, even though it currently boasts higher profit margins than HBM thanks to extreme scarcity, still cannot escape the gravitational pull of “fat profits attract expansion, expansion brings oversupply”; it remains cyclical.

So what you really need to watch is not “can prices notch new highs again,” but rather, when demand gets disrupted in some future quarter, whether gross margins will ease downward gracefully or plunge off a cliff.

Prices can deceive, inventories can deceive, but gross margins won’t – they are the one and only lie detector for whether this round is truly “super.”

If over the next several quarters, Micron’s gross margins can hold firm at elevated levels, retreating only gently even if demand fluctuates, then the “paradigm” camp wins. If margins peak and then collapse, then sorry – this is just another old cycle dressed in AI’s new clothes.

The second yardstick measures profit resilience, which depends on whether supply gets out of control.

And the biggest variable on the supply side is conspicuously absent from Micron’s earnings report. The third yardstick requires us to look beyond the report to find our answer.

The Third Yardstick: CXMT, the Variable That Didn’t Appear in the Earnings Report

The reason the memory industry holds such strong pricing power today rests on one word – oligopoly.

The global DRAM market is essentially carved up among three players: Samsung, SK Hynix, and Micron. Three players, much like three people co-running a casino; as long as no one upends the table by slashing prices in a production arms race, everyone gets to sustain high profit margins together. This “tacit restraint” is the invisible foundation beneath this cycle’s high gross margins.

But the one thing any oligopolistic structure fears most is the same thing – a fourth player showing up at the door.

For the memory Big Three, the biggest risk to pricing power has never been on the demand side, but on the supply side, in the form of that growing disrupter. That disrupter is called ChangXin Memory Technologies (CXMT), from China.

It did not appear in Micron’s earnings report, yet it could be the variable that determines the ceiling on Micron’s valuation.

And this variable is going through a critical shift in identity. In the past, when we talked about CXMT, the topic was “domestic substitution” – could they make it, could it be used, could imports be cut a little. But now, its story is moving from “pure substitution” toward “scale, profitability, high utilization.” These three words carry enormous weight: scale means its capacity is no longer piecemeal; profitability means it is no longer propped up by subsidies and has developed the ability to generate its own cash flow; high utilization means the production lines it has built are running at full capacity and the chips are genuinely selling. A competitor that can earn its own money and keep expanding is a threat of an entirely different magnitude compared to one surviving on life support.

Even more critical is capital. CXMT is already sprinting toward a STAR Market IPO, planning to raise nearly 30 billion RMB, and there is almost no mystery about where that money will go – continued capacity expansion. An IPO is not just about raising a round of funding; it is also a public “re-rating”: once it successfully lists and the market assigns an eye-catching market cap, CXMT will have a steady supply of ammunition to buy equipment, build fabs, and hire talent. The fourth player is no longer just standing far off at the door; it has already started knocking. The moment that door is pushed open, the rules of the high-margin game propped up by the “tacit understanding among the three” will have to be rewritten.

And its tangible progress in standard DRAM is in itself confirmation that “domestic competition has gotten real.” This is no longer a distant vision on a PowerPoint slide; it is yield rates and output that are coming off real production lines right now.

Let’s first lay out the current supply picture clearly, because hidden within it is a subtle chain reaction that is extremely favorable for price increases.

The Big Three are all doing the same thing right now: shifting production capacity toward HBM. Because HBM boasts high margins, everyone wants a bigger slice of the pie. But wafer capacity is limited — producing more HBM means producing fewer standard DRAM chips. This has created a counter-intuitive situation: the AI boom is actually causing the supply of ordinary memory to shrink. When standard DRAM is in short supply, prices also rise, so the price hikes have “spread” from the high end to every category. The ASP increases this quarter are the most direct footnote — DRAM average selling price rose by low-60s% quarter-over-quarter (just over 60%), while NAND even surged by mid-80s% sequentially (around 85%). This is precisely the underlying mechanism behind the current across-the-board memory price hikes, and we believe this supply-demand tightness will persist until after 2027.

Now, place CXMT on this chessboard.

CXMT's strategy is clear: avoid the most difficult high-end HBM for now, and instead enter through standard DRAM, rapidly scaling up with cost advantages and local demand. This creates a critical "time gap"—

• On the standard DRAM front, Chinese manufacturers are catching up faster than many expected. Once CXMT's standard product capacity is unleashed on a large scale, it could exactly fill the supply gap left by the Big Three as they shift production to HBM. The narrative that "AI squeezes out standard product supply and drives up prices across the board" would be weakened.
• But on the HBM front, the barriers are on a completely different level. HBM involves advanced packaging, stacking processes, and deeply integrated certification with AI chips — not something that can be quickly overcome just by throwing money and time at it. This high-end moat remains very deep in the near term.

So the third yardstick paints a "split" picture: CXMT's impact on standard DRAM is real and imminent, but its threat to high-end HBM is still a considerable distance away. This dovetails with the conclusions of our first two yardsticks — high-end is a paradigm, standard products are cyclical, and CXMT is the very catalyst that will cause the standard product cycle to "return early."

This also fills in an easily overlooked piece of the "super cycle" story: to judge how far this round can go, you can't just watch Micron and SK hynix's financial reports; you also have to watch that Chinese variable that doesn't issue US stock financial reports but is quietly reshaping the supply equation.

At this point, the significance of the long-term contract yardstick becomes even deeper.

Looking back at the first yardstick — why are customers in such a hurry to sign SCAs, pay advance payments, and lock orders until 2027? We used to interpret this as "locking price and quantity." But from a supply chain security perspective, there is a third meaning: locking capacity and security.

When geopolitics makes supply chains fragile, and when memory is seen as a "strategic resource" in the AI era, the largest customers want more than just a good price — they want the certainty that "no matter what happens, I have supply." Long-term contracts are essentially the insurance premium customers pay for supply chain security.

Behind those long-term orders that seem to lock in prices, what is truly being locked down is capacity, a sense of security, and the most expensive thing in an uncertain era — certainty.

This layer is exactly what the old cycle framework completely fails to capture. In the old framework, memory was a commodity, and the lowest-price supplier wins; in the new landscape, memory is a strategic resource, and whoever holds the supply is king. If this shift is real, its boost to pricing power is not a matter of just one or two quarters.

But we must also be honest — the supply chain security logic is a double-edged sword. While it elevates pricing power, it also spurs China to accelerate domestic substitution. The rise of CXMT and its peers is itself a product of geopolitical anxiety. In the short term, it helps the Big Three lock in customers; in the long run, it is also nurturing the very player that could flip the table.

The three yardsticks are measured. The duration of orders, the resilience of profits, the supply landscape — each points to the two-sided nature of the same conclusion. Now, let's return to that number that initially tempted you: 10x P/E.

The Valuation Trap: The Sweetness and Deadly Poison of 10x P/E

We've taken a long detour, and now we can directly answer the paradox from the beginning — why an explosive earnings report can trigger a sell-off.

First, let's throw out a number that will send a chill down your spine.

Over the past eight quarters, Micron has delivered eight earnings reports. In most of them, revenue and profit exceeded Wall Street expectations, and some were even "epic" beats. But after the reports, only twice did the stock manage to hold gains and deliver a positive return — the remaining six times, it either fell right away, or spiked in after-hours trading only to give back all the gains later.

Eight at-bats, two hits. A company whose revenue soared from $6.8 billion to $23.8 billion, with climbing performance, only saw its stock truly rise after earnings twice; the other six times it either fell outright or gave back initial gains. In other words, over the past two years, if you believed in the simple common sense that "good results should make the stock go up," you were probably slapped in the face by the market every quarter.

When it comes to Micron stock, "good earnings" and "good stock performance" are two things that have almost nothing to do with each other.

Don't believe it? Let's replay a few of the most typical scenes, and you'll understand what kind of cold, ruthless pattern lurks behind this.

Scene one: late 2024 (FY25 Q1). Revenue basically met expectations, and profit even slightly exceeded them — looking at the quarter alone, there was nothing to fault. But when earnings dropped, the stock cratered 16% the next day, its biggest single-day drop since March 2020. The trigger for the sell-off wasn't the "past" but the "future": management's guidance for the next quarter fell far short of market expectations and mentioned consumer electronics destocking. The results themselves were fine; the market feared that "the next quarter will get worse."

Scene two: mid-to-late 2025 (FY25 Q3 and Q4). Revenue set consecutive records, EPS massively beat, gross margin climbed from 39% to 46% — by any standard, this should have been a perfect report card. So what happened? The stock initially popped in after-hours (one spike up ~8%, another ~5%), but couldn't hold: either the next day gave back all the gains, even turning negative (one reversed into a nearly 3% loss), or it slowly drifted lower after the initial pop. The reason can be summed up in four words: sell the news. The stock had already risen in advance, and the moment earnings materialized, it became the starting gun for taking profits.

Scene three: late 2025 (FY26 Q1). Revenue surged 57% year-on-year, massively beating expectations, next-quarter guidance was stellar, and even capex was raised — this time, the stock finally rose, jumping ~10% the next day. This was the rare occasion among the eight where "good earnings matched good stock performance."

Scene four, the most extreme episode, was last quarter (FY26 Q2). Revenue nearly tripled year-on-year, gross margin hit 75%, EPS crushed estimates by multiples — the numbers were so beautiful they couldn't be more beautiful. Yet after earnings, the stock fell instead of rising, pulling back nearly 30% from its peak in about a week. But the truly magical part came in the second half of the story: if you didn't get shaken out during that panic, over the next three months the stock doubled from the rubble, eventually hitting new highs.

Stack these four scenes — actually eight quarters — together, and three genuinely useful patterns emerge.

First pattern, the most counterintuitive and systematic: good earnings ≠ good stock performance, and this is no accident, it's the norm. Out of eight times, six failed to close higher, and the worst drops often came with the most beautiful earnings numbers. The market is never buying "how much did they earn last quarter," but rather "can they keep beating expectations next quarter." What determines the stock price is always the gap between results and expectations, not the results themselves.

Second pattern, hidden in the pre-earnings price action: the more the stock outperforms the market before earnings, the harder it tends to fall afterward. The logic is not mysterious — if Micron significantly beats the market in the days before earnings (say, by more than 3%), it means the good news has already been front-run and priced in, leaving the report nothing but a "confirmation plus profit-taking"; conversely, if it has underperformed and no one was bullish, expectations are low, and the stock is more likely to rise after the report. This pattern is almost a reflex and worth checking every time before earnings.

Third pattern, the most useful for shareholders: first down, then up, is almost a fixed rhythm for Micron earnings. In the short term after earnings, the stock is more likely to fall, but stretched out over a month, it mostly rebounds. The most extreme version was last quarter — down 30% first, then doubled three months later. The short-term plunge is often not a collapse in fundamentals, but a clash of valuation frameworks and fear-driven shakeouts.

That big red candlestick after earnings usually doesn't kill the company's fundamentals, but your psychology. Once you understand "first down, then up," you won't hand over your chips in the darkest week.

It's worth noting that the triggers for these six drops (or failed rallies) varied: sometimes it was next-quarter guidance missing expectations, sometimes a sequential decline in gross margin that unnerved people, sometimes it was purely profit-taking after too big a run-up, and last quarter it was overvaluation combined with a rumor that "AI memory demand might not be as rigid as thought." The surface changes endlessly, but the core is one thing — the market is pricing "future expectations," not applauding "past results."

Does this mean earnings numbers don't matter at all? Not exactly. What truly ignites a sell-off is often a line that the market interprets as a "cycle peak" signal. And the most classic trigger of all is capital expenditure (capex).

Let's explain why this is a landmine. Capital expenditure is how much money a company plans to pour into building fabs, buying equipment, and expanding capacity. For an ordinary growth stock, increasing investment is a positive — it signals confidence in the future. But for a stock priced like a cyclical, raising capex is the loudest alarm bell. The market's knee-jerk reaction is: you expand capacity → a year or two later, overcapacity → price collapse → cycle peak. In the language of cyclicals, "I'm expanding capacity" roughly translates to "this cycle is about to end."

Thus we get that absurd yet logical scene: the more profit beats expectations, the more nervous the market; the more capex is raised, the fiercer the selling pressure. Smart money isn't selling "past goodness," it's selling "future overcapacity expectations."

This is the split personality of 10x P/E — to those who believe it's a growth stock, it's a sweet discount; to those who believe it's a cyclical, it's deadly poison bait.

So for this earnings, we need to fix our eyes on two key slots, which matter far more than the revenue number.

First: capital expenditure. This time management is clearly signaling an upward revision — FQ3 net capex of $7.1 billion, FQ4 expected to jump to around $10 billion, full-year FY2026 roughly $27 billion, higher than the previous expectation of just over $25 billion; even more worth noting, management also guided that quarterly capex in FY2027 will be even higher than FQ4, and more than half of the increase comes from construction spending. This is a real expansion ramp-up. By the old cyclical playbook, a large capex hike is the most jarring alarm, and even if earnings are stellar, you should be prepared for the market to sell off using the "cycle peak" logic. But from a different angle, most of this money is aimed at long-term contract capacity already signed — if the expansion "disciplinedly prioritizes SCA demand" rather than recklessly betting on spot prices, then the damage to the "paradigm" narrative isn't as severe. Disciplined, order-backed expansion is the evidence that the oligopoly's tacit understanding remains intact; the problem is, the market's knee-jerk reaction is almost always to price in "overcapacity" first.

Second: The initial stock price reaction after an earnings release is itself a vote of market sentiment — it tells you which yardstick the market is using at this moment. But this article was written right when the earnings first dropped, before the after-hours reaction could be fully absorbed. We can only say that if history rhymes, this time will likely see violent swings again.

Here’s a technical detail many people overlook that’s worth doing your homework on in advance: before its earnings report, Micron’s options implied volatility (IV) once spiked to 114%, and the market was betting real money on a single-day move as high as ±14%. Put into plain language — the market already knew it was a high-stakes gamble; no one was sure which direction the bullet would fly. And if you combine that with the three patterns mentioned earlier, you can even go one step further: first check whether Micron massively outperformed the broader market in the days before the report, or instead lagged behind — historical data shows that the more it outperforms, the harder it falls.

What does ±14% even mean? It means that whether it goes up or down, this was always destined to be a huge bullish or bearish candle. The market isn’t guessing whether “Micron is doing well” — everyone already knows it is. What the market is guessing is this: “Should this good performance be priced at 30x earnings like a growth stock, or at 10x like a cyclical stock?” That tug-of-war between valuation frameworks is the real source of the volatility.

So remember: after the earnings release, don’t just get caught up in the numbers game of revenue and EPS, and definitely don’t panic just because you see a giant bearish candle after the report. The real decisive hand is hidden in the tone around capex, management’s language on long-term agreements, and the vote the stock price casts with its feet. And history tells us that the short-term monster red candle is often just the first act of a “first down, then up” script.

A 10x P/E is not the answer — it’s a multiple-choice question. It forces you to answer what kind of company you really believe Micron is.

And to answer that question, we don’t need predictions; we just need to wait for a few signals to be verified.

Last Glow, or Paradigm Shift

At this point, you may want us to give you a clean-cut conclusion: buy, or don’t buy.

But we’re not going to hand you an answer; we’re handing you a set of coordinates. Because the question “last glow or paradigm shift” should never be resolved by guessing — it should be confirmed one by one through several verifiable signals.

We suggest you watch three of them closely. All are indispensable.

First signal: visibility into 2027.

If management and long-term agreements can clearly lock in orders all the way to 2027 or even further out, it means this business has truly transformed from a “spot stall” into an “annual subscription cafeteria” — the cycle has been re-anchored by long-term contracts. The longer the visibility, the stronger the paradigm flavor. If visibility still only extends one or two quarters, then deep down it’s still the same old cycle that lives and dies by the weather.

Second signal: the 80% gross margin threshold.

We view whether gross margin can break above and hold at elevated levels (80% is a symbolic watershed line) as the litmus test for profit resilience this cycle. Note: the key isn’t how high it can spike in a single quarter, but whether it holds when demand wobbles. If it holds, it’s a paradigm shift; if it crumbles at the first disturbance, it’s a last glow.

Third signal: capex discipline.

This is the most counterintuitive, and also the most important. If the three giants can collectively maintain production-expansion discipline — making money without blindly building capacity, instead serving long-term agreements with restraint and priority — then it shows the oligopoly’s “tacit understanding” is still intact, and high profits can be sustained. The moment someone loses patience first and capex surges across the board, the ending of this cycle has already been written — it’s only a matter of time.

Put these three signals together, and the decision logic becomes very clean:

If all three are confirmed at the same time — clear 2027 visibility, gross margin stabilizing at a high level, capex remaining restrained — then this cycle is a paradigm shift, and the 10x P/E is a bargain wrongly killed by the old framework. But if any single one of them collapses, it’s a last gasp, and that 10x P/E is a cup of sugar-coated poison.

That’s also why, from the very beginning, we’ve refused to slap any single label — “cyclical stock” or “growth stock” — onto Micron to price it. Right now it’s a stock living in the crack between two valuation frameworks — half its body has already stepped into the AI paradigm, while the other half is still soaking in the waters of the memory cycle. The market’s violent swings, the repeated spectacle of “good earnings matched with bad stock prices” six times over, are nothing but the result of those two frameworks tugging back and forth on it, again and again.

So, after those numbers get filled into the main body of the report, don’t be in a rush to cheer the revenue, don’t be seduced by the 10x P/E, and definitely don’t let the first candlestick after the earnings report set your rhythm. Go back to these three signals and check them off one by one.

When the tide rises, everyone looks like they’re swimming; only at the moment the tide goes out do you know whether the sea level has actually been permanently raised.