Semiconductor Configuration Logic After Micron’s Earnings: How to Choose Between HBM, DRAM, and NAND?

June 24, 2026 - Micron Technology delivered an earnings report that could be written into the history books of the memory storage industry after the U.S. stock market close. Revenue was $41.46 billion, up 345.7% year-over-year, exceeding market expectations by approximately 17.6%; Non-GAAP earnings per share were $25.11, far surpassing analysts' forecast of $20.28. This marks Micron's seventh consecutive quarter of earnings beats.

However, the timing of this report was quite delicate. On the same day during regular trading hours, Micron's stock price plunged 13.18%, closing at $1,048.51, with a trading volume of $63.37 billion. From the all-time high of $1,213.56 touched intraday on June 23 to the intraday low of $1,038.50 on June 24, the pullback exceeded 14%. The Philadelphia Semiconductor Index fell 7.87% on the same day, and the Nasdaq Composite Index closed down 2.21%.

The paradox of a "blowout" earnings report and a "plunging" stock price precisely provides the best entry point to understand the current investment logic of memory semiconductors. Based on the published earnings data, third-party institutional market estimates, and the industry's supply-demand structure, we attempt to establish a verifiable allocation analysis framework for three types of memory chips: HBM, DRAM, and NAND.

Key Data from Micron's Q3 Earnings: A Super Cycle of Volume and Price Growth

Let's look at a set of key numbers first. Micron's Q3 fiscal quarter revenue was $41.46 billion, up 73.75% quarter-over-quarter and 345.7% year-over-year. GAAP gross margin surged from 37.7% in the same period last year to 84.6%. The company's Q4 revenue guidance is between $49 billion and $51 billion, GAAP gross margin guidance of approximately 86%, and Non-GAAP EPS guidance of approximately $31.

From a product structure perspective, DRAM revenue reached a record $31.3 billion, accounting for 76% of total revenue; NAND revenue reached a record $9.9 billion, accounting for 24%. Data center-related business revenue increased nearly 7 times year-over-year, serving as the core engine for overall growth.

More critical information comes from the supply side. Micron revealed during the earnings call that the company's full-year 2026 HBM capacity is already fully sold out. The company expects memory supply tightness to continue beyond 2027, potentially improving gradually in 2028. Meanwhile, Micron has signed 16 long-term agreements with data center operators, automakers, and other customers, locking in sales over the next 3 to 5 years.

These data collectively point to a conclusion: the memory chip industry is in a structurally upward cycle driven by AI computing power demand, rather than a traditional inventory cycle rebound.

The Plunge Before the Earnings Report: What the Market Was Worried About

To understand the value of Micron's earnings report, one cannot ignore the market signals conveyed by the plunge before the release.

On June 24, Micron's stock plunged 13.18%. On the same day, SanDisk fell 13.64%, Western Digital fell 8.45%, and ARM fell over 10%. This sell-off was not an isolated event but the result of multiple pressures stacking up.

First pressure: the loosening of valuation anchors. According to historical data, when Micron's stock peaked in early 2022, its P/E ratio was only 9 times, and the stock later halved. At the cycle peaks in 1984 and 2018, P/E ratios were 15 times and 5.5 times, respectively. This pattern reveals a harsh reality: low P/E ratios for memory stocks often appear at cycle tops. After the earnings report, Micron's forward P/E based on future earnings is about just over 10 times – which seems "cheap" in traditional valuation frameworks, but in the context of the memory cycle, it is precisely a signal to be cautious.

Second pressure: the natural deceleration of growth. Before the Q3 earnings release, the market expected adjusted EPS to grow nearly 1,000% year-over-year. However, the year-over-year growth rate in the next fiscal quarter is expected to slow to approximately 725%. When the growth rate retreats from four-digit percentages to three-digit percentages, the market's logic for re-pricing valuations will fundamentally change.

Third pressure comes from the macro level. Hawkish signals from the Federal Reserve pushed the U.S. dollar index to around 101.5, near a 13-month high, while the 10-year U.S. Treasury yield remained around 4.5%. The discounting effect of high interest rates on future cash flows systematically suppresses high-valuation tech stocks.

The tension between the plunge and the earnings beat essentially reflects the market's differing views on "cycle positioning": is this an AI-driven structural paradigm shift, or a cyclical peak in the memory strong cycle?

HBM: The Most Certain Track Amid Supply Shortage

Among HBM, DRAM, and NAND, the supply-demand landscape for HBM is the clearest and offers the most allocation certainty.

From the demand side, the memory demand from AI servers is 8 to 10 times that of traditional servers. Major players like Nvidia continue to invest in AI data centers, accelerating their HBM procurement. South Korea's June memory semiconductor export data serves as a side confirmation – HBM (multi-chip packaging) exports surged 51% month-over-month.

From the supply side, HBM capacity from the three major memory manufacturers (Samsung, SK Hynix, Micron) is fully sold out. The industry's HBM capacity is approximately 330k wafers per month, potentially rising to 480k by 2027. But even with additional capacity coming online, Jefferies expects HBM prices could still rise by about 70% over the next 12 months.

From a pricing power perspective, Bernstein expects HBM prices could rise by 2 to 2.5 times by 2027. More noteworthy is the structural change – HBM pricing models are shifting from annual contract locking to dynamic repricing. Bernstein notes that since traditional DRAM's unit wafer revenue and gross margin are already significantly higher than HBM, memory manufacturers are renegotiating 2027 HBM prices with GPU/XPU makers. This means HBM's price elasticity is tilting in favor of suppliers.

The core logic for HBM allocation is a "supply rigidity × demand elasticity" double play. However, it is important to clarify that HBM investment opportunities are more reflected in the upstream memory manufacturers (Micron, SK Hynix, Samsung) rather than HBM itself as a tradable asset class. For readers in the crypto industry, HBM supply tightness has a pass-through effect on the cost structure of AI-related tokens and infrastructure projects – HBM price increases could raise AI computing costs, thereby impacting the economic models of AI sector projects.

DRAM: The "Spillover" Effect of Traditional Product Price Increases

DRAM is the most price-elastic category in this memory cycle and also the area with the greatest divergence among institutions.

Gartner data shows DRAM prices are expected to rise 125% in 2026. TrendForce data shows Q2 DRAM contract prices rose 58% to 63% quarter-over-quarter. Citi's forecast is more aggressive, expecting DRAM average prices to rise about 200% for the full year 2026. Bernstein data shows that from Q3 2025 to Q2 2026, traditional DRAM prices have cumulatively risen about 4.5 times.

The logic driving DRAM price increases differs from that of HBM. The concentration of wafer capacity towards HBM reduces supply for general-purpose DRAM, and its unit price has risen to 2 to 3 times that of the same period last year. In other words, DRAM's price increase is a "spillover effect" from HBM capacity crowding out, not a direct pull from AI demand. This "passive price increase" attribute determines that DRAM's price elasticity may be more cyclical than HBM's – once HBM capacity expansion is complete, the supply pressure on general-purpose DRAM will ease.

From a supply-demand gap perspective, excluding Chinese manufacturers, global memory bit supply growth is expected to be only 7% to 8% in 2026, with the combined supply gap for DRAM and NAND potentially reaching 150k to 200k wafers per month. Citi expects the global DRAM market to face a 5% supply gap in 2026.

The core logic for DRAM allocation is "price elasticity from capacity crowding out." However, caution is needed: DRAM's cyclical characteristics are more pronounced than HBM's. Currently, traditional DRAM's revenue per wafer is about twice that of HBM, and gross profit is nearly three times higher. This extreme distortion in profitability is incentivizing memory manufacturers to shift more capacity towards traditional DRAM, and this capacity rebalancing could become a turning point signal for the DRAM price cycle.

NAND: Highest Upside but Highest Risk

Among the three types of memory chips, NAND has the highest expected price increase for 2026, but its supply-demand structure is also the most fragile.

Gartner data shows NAND prices are expected to rise 234% in 2026. TrendForce data shows Q2 NAND flash contract prices surged 70% to 75% quarter-over-quarter. Citi expects NAND to rise about 186% for the full year. Some institutions are even more aggressive in their forecasts for Samsung Electronics' NAND ASP, expecting a 283% year-over-year increase in 2026.

NAND's demand pull also comes from the expansion of AI infrastructure – the buildout of AI inference servers drives explosive demand for NAND and SSDs, with both categories growing 25% to 28% quarter-over-quarter. In South Korea's June memory semiconductor exports, the month-over-month growth rates for NAND and SSDs both remained above 25%.

However, NAND faces two structural risks.

First, a more fragmented competitive landscape. Unlike the highly concentrated market structures of DRAM and HBM, the NAND market has more participants, including Samsung, SK Hynix, Western Digital, and KIOXIA. Bernstein maintains an "underperform" rating on KIOXIA, which lacks an HBM business, reflecting the market's pricing discrimination against pure NAND players.

Second, the potential threat from Chinese manufacturers. Jefferies reports that while Chinese NAND manufacturers are unlikely to threaten global leaders in the short term, by 2028, Chinese NAND technology may become more globally competitive. This means NAND's long-term supply landscape faces greater uncertainty.

The core logic for NAND allocation is a "high elasticity × high risk" hedging combination. For investors seeking high returns, NAND's upside elasticity offers the greatest imagination space; but for risk-averse allocations, NAND's competitive landscape and long-term supply uncertainty make it the most uncertain risk-adjusted return among the three product categories.

Allocation Framework: Differentiated Positioning for Three Product Types

Based on the above analysis, the allocation logic for HBM, DRAM, and NAND can be summarized into three different investment paradigms:

HBM corresponds to a "certainty premium." Full-year capacity sold out, long-term agreements locked in, and pricing model renegotiation – these three barriers make HBM the category with the strongest supply rigidity and highest demand visibility among the three. Allocating to HBM is essentially allocating to the "pickaxe seller" logic of AI computing infrastructure, rather than betting on price cycle elasticity.

DRAM corresponds to "cyclical elasticity." Passive price increases from capacity crowding out, a 5% supply gap, a cumulative 4.5x price increase – these data point to a typical memory upcycle. But DRAM's cyclical nature means it can be both the most elastic and the most prone to sharp pullbacks. Allocating to DRAM requires judgment of cycle positioning.

NAND corresponds to "high risk, high reward." The expected 234% annual price increase is the highest among the three categories, but the fragmentation of the competitive landscape and long-term threats from Chinese manufacturers are also the largest. Allocating to NAND is more suitable as a "satellite position" in a portfolio rather than a core holding.

It is particularly important to note that the above analysis framework is based on a key premise: there will be no large-scale expansion of global memory capacity between 2026 and 2027. Jefferies expects no significant wafer capacity growth in 2027. However, if global wafer capacity grows 15% to 20% in 2028, coupled with a slowdown in AI demand, memory prices could see a sharp decline. Bernstein also warns that a cyclical downturn is still possible in 2028. This time window constitutes the "hard constraint" for all memory allocation strategies.

Conclusion

The significance of Micron's Q3 earnings goes beyond setting company records for revenue, gross margin, and cash flow. It provides a clear demand coordinate system for the entire memory semiconductor industry – AI data centers' appetite for HBM far exceeds market expectations, and the spillover effect of this demand is reshaping the pricing logic of DRAM and NAND.

However, the fact that the stock price plunged 13.18% on June 24 reminds us: even the strongest fundamentals must face the gravity test of valuations. The semiconductor sector's rolling P/E ratio has exceeded 210 times, in the 99th percentile of history. When the entire sector's valuation is near historical extremes, any marginal negative information can trigger sharp valuation re-pricing.

For investors, the real question after Micron's earnings may not be "whether to buy memory," but "at what point in the cycle to buy, and which type of memory to buy." HBM's certainty, DRAM's elasticity, and NAND's high reward – three product types correspond to three different risk appetites and holding periods. Before the potential cycle turning point of 2028, the allocation value of the memory track remains, but the ability to select products will be more important than ever.

FAQ

Q1: What was the most surprising data in Micron's Q3 earnings?

Micron's Q3 revenue was $41.46 billion, far exceeding market expectations of $35.25 billion by approximately 17.6%, up 345.7% year-over-year. Non-GAAP EPS of $25.11 also significantly beat the market estimate of $20.28. This was Micron's seventh consecutive quarter of earnings beats.

Q2: What are the expected price increases for HBM, DRAM, and NAND in 2026?

According to Gartner, DRAM prices are expected to rise 125% and NAND prices 234% in 2026. For HBM, Bernstein expects prices could rise 2 to 2.5 times by 2027. TrendForce data shows Q2 DRAM contract prices rose 58%-63% quarter-over-quarter, and NAND rose 70%-75%.

Q3: Why did Micron's stock price plunge despite the earnings beat?

Before the June 24 earnings release, Micron's stock had already plunged 13.18%. Core reasons include: market doubts about the sustainability of the AI memory cycle, hawkish Fed signals suppressing tech stock valuations, and the historical pattern of "low P/E ratios often appearing at cycle tops" in the memory industry causing concern. The earnings beat was a "good news realization," and the plunge occurred before the earnings release.

Q4: Among the three types of memory chips, which is the most worth allocating to?

The three product types correspond to different risk appetites: HBM has the strongest supply rigidity and highest demand visibility, suitable for allocations seeking certainty; DRAM has the greatest cyclical elasticity, suitable for investors with judgment on cycle positioning; NAND has the highest expected annual price increase but the most fragmented competitive landscape, suitable as a satellite position in a portfolio. Note the potential cycle downturn risk in 2028.

Q5: How does the memory chip price increase impact the crypto industry?

Memory chip price increases impact the crypto industry through two main channels: first, higher computing costs for AI-related tokens and infrastructure projects, potentially compressing project economic models; second, volatility in memory semiconductors as risk assets transmits to the crypto market through market sentiment. On June 24, Bitcoin fell to $59,018, correlating with that day's tech stock sell-off.