Author: Godot
The storage oligopoly boom and operating leverage explosion during the HBM cycle
Treating Micron as a storage company is the biggest misconception in valuation.
The market's current perception of Micron remains within the framework of a traditional cyclical stock, severely underestimating the fundamental difference between this AI-driven cycle and historical cycles.
This article will analyze Micron's profit revaluation logic from three dimensions: industry physical structure, oligopolistic competition discipline, and pricing mechanism.
The fundamental difference between DRAM and NAND: Capital expenditure efficiency determines the depth of the competitive advantage.
Treating Micron as a storage company is the biggest misconception in valuation.
DRAM and NAND differ fundamentally in their physical structure, technological barriers, and return on capital.
The core memory unit of DRAM is the capacitor, which is used to store electrical charge. As the manufacturing process has progressed from 1-alpha to 1-beta, and then to the current 1-gamma node, the size of capacitors has continued to shrink, and leakage problems have become increasingly difficult to solve, forcing manufacturers to introduce EUV extreme ultraviolet lithography machines, which cost hundreds of millions of dollars per unit.
This means that even if second-tier manufacturers invest heavily, it will be difficult for them to catch up with the technological advantages of the leading players.
NAND is completely different. 3D NAND expands capacity by vertically stacking storage cells, and the industry is currently moving from 232 layers to over 300 layers.
While this technological approach also consumes capital, the barriers to entry are relatively manageable. Second-tier manufacturers such as Kioxia and Western Digital can relatively easily increase supply by investing capital in purchasing stacking equipment.
This is the fundamental reason why NAND's gross margin has been lower than DRAM's for a long time and why it is more prone to price wars.
Micron's current revenue structure consists of approximately 70% DRAM and approximately 30% NAND. In SOTP valuation, the DRAM business should rightfully enjoy a higher valuation multiplier, while the NAND business should be given a discount.
The core rationale for buying Micron is to capitalize on the oligopoly advantage in DRAM and the growth potential of HBM.
Oligopoly: The True Anchor of Valuation
The storage industry has undergone a brutal process of consolidation, from dozens of players to just three remaining: Samsung, SK Hynix, and Micron.
The key variable for judging the health of an industry is not the battle for market share, but capital expenditure discipline (CapEx Discipline). When oligopolies frantically expand production to grab market share, the entire industry suffers losses. Conversely, when supply is tacitly controlled, profits amplify exponentially.
After experiencing the previous deep downturn, the three giants have shown unprecedented restraint on the supply side. A more crucial variable is the capacity-consuming effect of HBM, which constitutes the biggest structural difference between this cycle and previous cycles.
HBM capacity squeeze effect: Underestimated supply-side shortage
The market generally knows that HBM has high prices and good gross margins, but it underestimates the extent to which HBM squeezes traditional DRAM production capacity at the physical level.
This mechanism is composed of two factors:
1) Die Size Penalty : HBM consists of multiple layers of DRAM dies (HBM3E is typically 8 or 12 layers) plus a layer of basic logic die. To achieve ultra-high bandwidth, a large amount of wiring space needs to be reserved inside the chip. For the same capacity, an HBM chip occupies approximately 2 to 2.5 times the silicon area of a traditional DDR5 chip.
2) Yield Loss: HBM relies on through-silicon vias (TSVs) to drill through multiple layers of bare dies and connect them with copper pillars. If any layer is misaligned or the drilling fails, the entire HBM die is scrapped. This makes the overall yield of HBM significantly lower than that of ordinary DRAM.
The combination of these two factors means that producing one HBM wafer is equivalent to consuming the capacity that could have produced three to four standard DDR5 wafers. Even if demand from PCs and smartphones recovers slowly, as long as the three giants allocate their best capacity to Nvidia for HBM production, the supply of traditional DRAM will be artificially compressed, thereby supporting and pushing up the average selling price across the entire industry.
Therefore, as long as HBM demand does not experience a precipitous decline, Micron's pricing power in traditional DRAM remains extremely secure. The market should not easily predict a repeat of a price war.
Pricing Mechanism Reassessment: From Bit Growth Driven to ASP Driven
Micron's basic revenue formula is: Revenue = Total Shipments × Average Selling Price (ASP).
Historically, the boom in storage cycles has been driven by bit growth and ASP double-digit growth, a typical scenario being the smartphone boom.
However, this cycle is characterized by a completely different pattern. Bit growth has remained at a moderate level of only mid-to-low single digits, while the overall ASP has risen sharply due to the inclusion of HBM. HBM is typically priced several times higher than standard DRAM and is locked in through prepayment mechanisms and long-term agreements.
In the past, Micron's performance fluctuated wildly with the spot market, but now a significant proportion of its capacity is locked in high-margin, long-cycle AI orders. The peak gross margin of this cycle is expected to surpass historical extremes.
Operating leverage: Non-linear profit surge under a fixed cost structure
Semiconductor manufacturing is a typical capital-intensive industry with high fixed costs. An advanced wafer fab can cost between $15 billion and $20 billion, and equipment depreciation is typically completed within 5 to 7 years.
Assuming Micron's base cost for producing a batch of wafers is $100, of which $40 to $50 is fixed equipment depreciation, which must be accrued regardless of production volume. The remaining variable costs, such as materials and labor, account for about half.
When the ASP rises from $100 to $120, since fixed costs remain unchanged, almost 100% of the additional $20 is converted into operating profit. This is the power of operating leverage.
The reverse logic also applies: during a downturn, once the ASP (Average Selling Price) falls below the fixed cost line, the company will face severe cash flow problems. This is the fundamental reason why the stock price volatility of storage stocks far exceeds that of fundamental fluctuations.
summary,
When constructing Micron's financial models for the next few quarters, assumptions about the rate of gross margin expansion should be more aggressive than those of traditional cyclical models. HBM locks in high-margin capacity, oligopolistic discipline suppresses price competition, and operating leverage amplifies the profit elasticity of rising ASP.
Micron's valuation represents a structural reassessment of its profitability. While the market is still pricing Micron within the traditional price-to-book ratio framework for cyclical stocks, this presents a window of opportunity for institutional investors to build positions.
Of course, the core risk variable that needs to be continuously monitored is the demand for HBM endpoints. Once cloud vendors slow down their capital expenditures or Nvidia's GPU shipment pace reaches an inflection point, the fulcrum of the entire logic chain will loosen.
CapEx, technology pace and HBM packaging bottlenecks
CapEx game drives up stock price
In most industries, management cutting capital expenditures is interpreted as a loss of confidence in the future. But the logic for storage is the opposite.
Due to the industry's strong cyclical nature, the ratio of capital expenditure to revenue is a core indicator for judging whether the supply side is out of control.
A healthy level is usually around 30%. Once it exceeds 40%, it means that the entire industry has entered a frenzied expansion mode, and will inevitably face overcapacity and bloody price wars in the future.
During the super downturn of 2022-2023, Micron's management decisively cut wafer fab equipment spending by nearly 50% in fiscal years 2023 and 2024.
Wall Street paid a very high premium, and cutting CapEx not only protected the bottom line of free cash flow, but also sent a clear signal to the market that it was maintaining its value.
When professional analysts see such announcements, they will not lower their target prices; instead, they will regard them as the first buy signal that the cycle has bottomed out and the stock price has reversed.
Alternatively, CapEx can be split into two parts: the factory and the wafer fabrication equipment.
There is a significant time lag between these two types of expenditures. Building a wafer fab takes 2 to 3 years, while moving in and debugging equipment such as lithography machines only takes 6 to 12 months.
Micron cut equipment spending by 50% during the downturn, directly reducing short-term production and supporting market prices; however, plant construction spending, such as the upfront investment in new plants in Idaho and New York, was not significantly reduced.
This strategy avoids disrupting supply and creating price wars in the short term, while reserving physical space for the next AI supercycle in the long term.
Once the demand inflection point appears, Micron can quickly purchase equipment to fill its factories and seize market share. This is a typical counter-cyclical investment strategy.
Micron's technology roadmap
Precise timing of the 1-gamma node
The technological approach is essentially a matter of ROI. An ASML EUV lithography machine costs hundreds of millions of dollars and consumes an astonishing amount of electricity.
If the cost savings from technological upgrades cannot cover the huge depreciation of new equipment, it will be financially devastating.
Micron has taken a completely different path from Samsung in the past few years. At the 1-alpha and 1-beta nodes, Micron pushed traditional DUV technology to its limits, achieving mass production earlier than Samsung, while cleverly delaying the full-scale introduction of EUV.
As a result, Micron's manufacturing costs at these two points were actually lower than those of Samsung, which was the first to introduce EUV, demonstrating its leading capital efficiency among its peers.
With the advent of the current 1-gamma node, the physical dimensions have shrunk to the point where the cost of DUV multiple exposures exceeds that of using EUV directly. It was only then that Micron began to fully and maturely introduce EUV equipment, striking the perfect balance between cost reduction in technology and equipment depreciation.
In our financial model, we expect Micron's 1-gamma node to have a smoother yield ramp-up than when Samsung first introduced EUV, which will directly translate into better gross margin performance.
HBM3E overtaking on a curve
In the AI era, the ability to obtain NVIDIA certification determines a storage vendor's pricing power.
The HBM field has long been dominated by SK Hynix, while Micron skipped HBM3 and directly poured all its R&D resources into the more advanced HBM3E.
It has successfully entered the supply chain of NVIDIA H200 and B100, and its power consumption is nearly 30% lower than that of its competitors.
In data center scenarios, power consumption directly determines cooling costs and is a core metric for hyperscaler selection. As a result, Micron's HBM production capacity for 2024 and 2025 has been completely booked.
This change has fundamentally impacted valuation methodologies. HBM businesses are no longer suitable for pricing using the traditional price-to-book ratio framework for cyclical stocks; instead, they should be priced using the price-to-earnings multiple of growth stocks within the SOTP framework. This is the core driver behind Micron's valuation restructuring.
Bottleneck: HBM isn't created, it's pieced together.
The market generally knows that HBM is in short supply, but the real bottleneck in production capacity is not in the front-end silicon wafer manufacturing, but in the back-end advanced packaging.
The core process of HBM is to stack 8 or 12 layers of DRAM dies as thin as cicada wings together, and then use through-silicon via (TSV) technology to drill tens of thousands of tiny holes and connect them with copper pillars.
Any misalignment or heat dissipation issue can render the entire expensive HBM chip unusable, making the yield and capacity expansion of HBM highly dependent on the maturity of the packaging process.
More importantly, there is the parasitic relationship within the supply chain. Micron's HBM3E chips cannot be delivered directly to end customers; they must be sent to TSMC and integrated with Nvidia's GPUs onto a single packaging substrate via CoWoS.
This means that Micron's HBM revenue growth is firmly locked onto TSMC's CoWoS capacity expansion curve.
When assessing Micron's HBM revenue, one cannot only look at Micron's own capacity planning; it is also necessary to keep a close eye on the expansion speed of TSMC's CoWoS capacity.
If TSMC's packaging process stalls, Micron's output of HBM will not translate into revenue in the current quarter. This is one of the most important cross-validation indicators for analysts tracking the entire AI computing power supply chain.
summary,
The tacit agreement among DRAM oligopolies maintained the price floor, the counter-cyclical CapEx strategy preserved cash flow, the precise timing of the EUV technology route improved capital efficiency, and the HBM3E generational bet opened up the valuation ceiling.
We will focus on four high-frequency indicators: the split ratio of WFE and Shell in Micron's quarterly CapEx, the yield ramp-up progress of the 1-gamma node, the changes in the allocation of Micron's HBM share in Nvidia's supply chain, and TSMC's CoWoS monthly capacity data.
A turning point in any of these four indicators could trigger a reassessment of the target price. Among them, TSMC's CoWoS capacity is the most easily overlooked yet most directly influential variable in determining Micron's near-term revenue realization, and it requires close monitoring.
Micron's Cyclical Investment Indicator System
When a storage company delivers its most lucrative financial report in history and its price-to-earnings ratio is incredibly low, it's often the peak selling point; conversely, when a company reports huge losses and its profit and loss statement is dismal, it's an excellent time to buy.
The root of this counterintuitive phenomenon lies in the fact that stock pricing is a forward discounting of the fundamentals over the next 6 to 9 months.
Since historical financial reports are no longer useful, what forward-looking indicators do institutional analysts rely on to judge turning points?
Inventory turnover days (DIO)
For memory chips, inventory is poison.
Chips not only depreciate in value over time, but also tie up a significant amount of working capital. The balance sheet metric most closely watched by institutional analysts is DIO (Days Inventory Outstanding).
Historical experience shows that Micron Health's DIO (Distribution Intake) ranges from 90 to 110 days. Once DIO exceeds 130 days and continues to climb, it means that the produced chips cannot be sold and are stuck in warehouses.
Even though the quarterly profits look decent, a price war is imminent, and the rating must be downgraded immediately.
But what deserves more attention are the bottoming signals.
At its worst in 2023, Micron's DIO lifespan soared to over 200 days.
The real buying opportunity is not when DIO returns to a healthy 100-day level, but when DIO starts to decline sequentially from its 220-day high, for example, when it drops to 180 days.
This means that destocking has begun and the fundamentals have bottomed out. By the time DIO returns to normal levels and you enter the market, the main upward trend in the stock price will have already ended.
You can't just look at Micron's own warehouses.
Once the chips are sold from Micron to customers such as Dell, Apple, and Amazon AWS, they disappear from Micron's balance sheet and are included in revenue.
But this doesn't mean the chip was actually consumed; it may simply have been moved from Micron's warehouse to the customer's warehouse.
Institutional analysts build massive channel tracking models, scrutinize the financial reports of PC manufacturers such as Dell, HP, and Lenovo, as well as major cloud service providers, to reverse-engineer the number of weeks of customer component inventory.
In the early stages of a downturn, end-user demand begins to weaken, but customers continue to order due to long-term contracts. At this time, Micron's financial reports still look good, but customer warehouses are already overflowing. If a major customer suddenly announces a halt to deliveries and the need to clear inventory, Micron's revenue will instantly be halved.
A true bottoming signal requires two conditions to be met. Micron's own inventory begins to decline, and channel customers' inventory simultaneously drops to normal levels, typically within 4 to 6 weeks.
The real starting point of Micron's stock price surge is when major customers' inventories run out and they are forced to return to the market to start a restocking cycle.
Is inventory impairment a negative or a positive factor?
The core issue is the LCM (lower of cost or net realizable value) principle in accounting. The logic is as follows: Suppose Micron's cost to produce a chip is $100. If an industry downturn causes the market price to drop to $80, Micron must immediately accrue a $20 inventory impairment loss according to accounting standards. This would cause the quarterly gross margin to collapse instantly, or even turn into a deep negative figure.
The book cost of these chips has been artificially lowered to $80. When the market recovers and prices rise back to $95 in the next quarter, Micron will show a profit of $15 on its financial statements, not a loss, when it sells this inventory.
When Micron announces massive inventory impairment during earnings calls, retail investors often panic and sell off their shares, but institutional investors know that this is management cleaning up the financials.
Once the burden is shed, the gross margin elasticity in the following quarters will be significantly amplified. This is precisely the accounting basis for the phenomenon of stock prices rising as soon as negative news is fully priced in at the beginning of a cyclical reversal.
The income statement is fake; the cash flow statement is the real one.
Inventory impairment can lead to a collapse in gross margin and a sharp drop in net profit, but this does not mean that Micron will go bankrupt. This misconception is the fundamental reason why short sellers at the bottom of the cycle are repeatedly wiped out.
Inventory impairment is essentially a non-cash expenditure; it is merely an adjustment to the book value and does not involve any actual cash outflow.
More importantly, during the industry's darkest times, Micron will significantly reduce raw material purchases and stop accumulating inventory. Changes in accounts payable and inventory will actually generate positive cash inflows in the cash flow statement. This is the reverse release effect of working capital.
During periods of extreme economic downturn, institutions completely disregard Micron's P/E or EPS figures because they are all negative and therefore meaningless.
We should keep a close eye on free cash flow (FCF). If Micron suffers a net loss of $2 billion in the quarter, but its free cash flow miraculously remains positive or only slightly declines, it indicates that management has stopped the bleeding through operational measures.
Cash flow always bottoms out one to two quarters earlier than profits, making it one of Wall Street's most profitable leading indicators.
When DRAM and NAND flash memory prices plummeted due to overcapacity, Android phones that originally came standard with 8GB of RAM saw their prices halved. Phone manufacturers discovered that upgrading to 12GB or even 16GB of RAM not only didn't increase costs significantly, but also served as a major marketing selling point. The result was a sudden surge in the amount of RAM installed in individual phones.
This doubling of single-unit capacity triggered by a price crash will quickly absorb excess inventory in the market.
When industry chain research shows that mid-to-low-end mobile phones and ordinary PCs begin to be equipped with large-capacity memory on a large scale, the bit crossover point has arrived, and the destocking cycle will end at a speed far exceeding expectations.
In-depth analysis of the Q2 FY2026 financial report: When a P/B ratio of 8 becomes the new normal
Total quarterly revenue was $23.86 billion, a surge of nearly 200% compared to $8.05 billion in the same period last year.
The DRAM business contributed the vast majority of revenue, reaching US$18.768 billion, a year-on-year increase of 207%.
What deserves more attention than the revenue figures is the internal structure of revenue growth.
MD&A’s financial report clearly states that the significant increase in DRAM revenue mainly came from a mid-range increase in average selling price (ASP) of 60% quarter-on-quarter, while bit shipments only saw mid-single-digit growth.
This is direct evidence of a complete victory in pricing power. Micron doesn't need to sell more chips to make money; HBM's structural shortages give it the power to raise prices.
SOTP Valuation
The sum-of-the-parts SOTP valuation is crucial, especially the data from the two core business units (BUs) disclosed in Micron's financial statements.
1) Cloud storage services
Q2 revenue was $7.749 billion, nearly tripling from $2.947 billion in the same period last year.
Even more alarming is the operating profit margin, which is as high as 66%.
Micron has seized the pricing power for HBM and high-capacity DDR5 for data centers.
2) Mobile and Client Services
Traditionally, competition in the mobile phone and PC memory markets has been fierce, putting long-term pressure on gross profit margins.
However, in Q2, the business generated $7.711 billion in revenue, compared to only $2.236 billion in the same period last year, and its operating profit margin reached an unbelievable 76%, compared to only 47% in the same period last year.
After HBM drained the entire industry's wafer capacity, the capacity squeeze effect led to a severe structural shortage of memory for ordinary mobile phones and PCs. Micron also achieved unprecedented profits in its traditional business.
This is the product of the combined resonance of cyclical and structural shortages.
Gross profit margin hits record high
The Q2 non-GAAP gross margin reached 74.9%, and the Q3 guidance is even higher at around 81%.
This level of gross margin is typically found only in pure software companies like Microsoft, or in AI chip design companies that hold an absolute monopoly.
For a capital-intensive wafer manufacturing company to deliver such high-quality profitability data is something the memory industry has never seen in the past 30 years.
Discrepancy between inventory and accounts receivable
The divergence between inventory and accounts receivable signals a very strong seller's market.
Despite a 75% quarter-over-quarter increase in Q2 revenue, ending inventory was only $8.267 billion, even slightly lower than the $8.355 billion at the end of the previous fiscal year.
The chips were all taken away almost immediately after being stored in the warehouse.
Accounts receivable nearly doubled, soaring from $9.265 billion at the end of the previous fiscal year to $17.314 billion at the end of Q2.
It's not that customers are defaulting on payments, but rather that the speed of shipments and the magnitude of price increases are too fast. Giants like Nvidia and Microsoft are frantically taking delivery, and a huge amount of payment is still within the 30 to 60-day payment period.
This foreshadows an extremely strong positive surge in operating cash flow next quarter. Inventory turnover days (DIO) are at an extremely healthy low, with no risk of oversupply.
In Q2, net cash provided by operating activities reached $11.9 billion, net capital expenditures excluding government subsidies were $5 billion, and adjusted free cash flow reached $6.899 billion.
Management also announced a 30% increase in the quarterly dividend, a clear signal of extreme confidence in future cash flow.
However, capital expenditures (excluding government subsidies) are projected to exceed $25 billion in fiscal year 2026. Micron is aggressively expanding production in Idaho, New York, and even at its newly acquired wafer fab in Taiwan for $1.8 billion.
This means that a large portion of the massive cash Micron earns will immediately be reinvested in equipment purchases and factory construction. It's a double-edged sword.
Based on the Q2 FY2026 financial report, the quarterly revenue was US$23.86 billion, the non-GAAP diluted EPS was US$12.20, the Q3 gross margin guidance was approximately 81%, the diluted outstanding shares were approximately 1.142 billion, and the current net asset value per share was approximately US$64.2.
Over the next 6 to 12 months, corresponding to the second half of FY2026 and FY2027, the market will be entirely dominated by the supply shortage of HBM and software-level profit margins.
Wall Street has completely abandoned the traditional price-to-book (P/B) model and adopted the sum-of-the-parts (SOTP) and high-growth P/E model to price Micron.
Micron is no longer a memory chip manufacturer, but a core player in AI computing infrastructure, enjoying the valuation premium of Nvidia.
According to management guidance, Q3 EPS will reach $19.15, with a fluctuation of $0.40.
Extrapolating from this momentum, annualized EPS will easily exceed $75. This level of profitability is unimaginable in the traditional storage industry.
Under the SOTP framework, AI/HBM-related businesses are given a forward P/E multiplier of 20 to 25 times (currently 27.22), while traditional businesses are given a high P/B multiplier of 2.5 times.
The overall goal is obtained by summing the two parts.
- Target price: $1,300 to $1,500
- Target market capitalization: $1.48 trillion to $1.71 trillion
This will be the absolute pinnacle in the history of memory semiconductors.
Retail investors and trend-following CTA funds will flood in, but institutional funds will respond by implementing strict, phased reductions in their positions as the stock price approaches $1,400.
The core judgment here is that when everyone is using the P/E ratio of growth stocks to price cyclical stocks, the valuation anchor is already suspended, and it is only a matter of time before it breaks down.
Any of the four major cloud giants has hinted that the return on AI investment is lower than expected; HBM spot prices have begun to ease slightly; and TSMC's CoWoS capacity expansion is outpacing Nvidia's demand growth.
The appearance of any one of these three signals should trigger a reduction in holdings; the simultaneous appearance of all three signals indicates the end of Phase One.
