Produced by I Xiahaifallsea Written by I Hu Buzhi

After the market closed on August 5 (Eastern Time), Western Digital (WDC) released its financial results for the fourth quarter of fiscal year 2026. Revenue reached $3.75 billion, up 44% year-over-year, hitting the upper end of the guidance range and surpassing market expectations of $3.69 billion. GAAP net profit was $3.195 billion, up more than 12 times year-over-year. Adjusted earnings per share were $3.56, higher than analysts' expectations of $3.31. GAAP gross margin was 54.1%, up 13.1 percentage points year-over-year. Next quarter's guidance: median revenue of $4.1 billion, gross margin of 55% to 56%, both exceeding market expectations.

Revenue exceeded expectations, profit exceeded expectations, gross margin exceeded expectations, and guidance exceeded expectations. The combination of four "exceeding expectations" is rare in any company's quarterly financial report.

But after the market closed, Western Digital's stock fell more than 11%.

A financial report showing a 12-fold increase in profit resulted in a double-digit drop in stock price. This seems contradictory at first glance, but the logic becomes clear when looking at Western Digital's stock performance over the past year. As of August 5, its year-to-date cumulative gain exceeded 200%. Over the past 12 months, it has surged more than 600%. The stock price climbed from around $50 to a peak of nearly $800. The market had already factored in all the positive factors, such as AI storage demand, HDD supply shortages, and margin expansion, into the price.

At this level, the term "exceeding expectations" is no longer sufficient. The market doesn't just want "better than expected"—it wants "better than what's already priced in." But if it stopped there, this would just be a conventional analysis of "profit-taking."

What truly deserves discussion is a deeper issue. The yardstick the market uses to measure Western Digital is that of a "cyclical stock." Yet Western Digital is becoming a non-cyclical company. Measuring compounding with a cyclical yardstick naturally leads to the conclusion that it has "peaked." Using a different yardstick reveals a completely different picture.

01

First, let's examine where Western Digital excelled this quarter.

Revenue reached $3.75 billion, up 44% year-over-year. The number itself is impressive, but more important is its composition. Cloud business revenue for the quarter was approximately $3.3 billion, accounting for 89% of total revenue, up 43% year-over-year. Nearly all growth came from data centers and hyperscale customers.

This means Western Digital is no longer a "PC hard drive company that happens to do some enterprise business"—it has become a company with AI infrastructure storage as its core revenue source. Gross margin was 54.1%, up 13.1 percentage points year-over-year. This improvement wasn't achieved by "selling more." This quarter, Western Digital shipped 231EB of storage capacity, up 22% year-over-year. However, the blended average selling price per TB rose by a "high teens" percentage year-over-year.

Last quarter, this figure was only a "high single digits" percentage. With a 22% increase in volume and a "high teens" increase in price, revenue grew by 44%. Price contributed more to growth than volume. This indicates strengthening pricing power. It's not "trading volume for price"—it's "volume and price rising together." In the semiconductor and storage industry, this typically only happens during periods of extreme supply-demand tightness. Adjusted operating margin was 44.2%, up 16.1 percentage points year-over-year. For every dollar earned, 44.4 cents was profit. Free cash flow margin was 34%.

Capital expenditures were only $108 million, accounting for less than 3% of revenue. Compared to AI chip companies, which often have capex ratios of 20% to 30%, Western Digital requires almost no additional capital investment to achieve growth. This represents extremely rare capital efficiency. GAAP net profit was $3.195 billion, up more than 12 times year-over-year. This figure includes one-time factors, but even excluding them, adjusted net profit still reached $1.382 billion, more than doubling year-over-year.

Next quarter's guidance calls for revenue of $4.1 billion, up about 45% year-over-year. Gross margin is expected to be 55% to 56%. Management stated on the earnings call that full-year FY2026 revenue will grow 36% to $12.9 billion, with earnings per share doubling. CEO Irving Tan used a notable phrase on the call.

He said the company is in the "closing phase of its first full fiscal year as a focused HDD company." This wording implies a shift in identity. Western Digital no longer defines itself as a "storage company" but as a "focused HDD company." After completing the spin-off of its flash memory business in February 2025, the company has concentrated all its resources on mechanical hard drives. These numbers together paint a very clear picture. Western Digital's fundamentals have never been this strong.

But the 11% after-market drop is also real.

02

To understand this drop, we must first understand the underlying logic of how the market prices Western Digital.

For the past two decades, the HDD industry has been perceived in the capital markets as a classic "cyclical stock." Demand fluctuates with macroeconomic cycles and data center construction cycles. After a boom, demand falls, prices drop, and gross margins contract. The downturn from 2022 to 2023 is the most direct evidence. Western Digital's gross margin fell from 39% to around 25%, and its stock price halved. The entire industry underwent brutal destocking.

So, when gross margin surged from 25% to 54%, the instinctive reaction under the "cyclical stock" framework is: It's probably peaked.

This reaction isn't unreasonable. The valuation logic for cyclical stocks is "mean reversion." After a peak, a trough must follow. When a cyclical company's profitability reaches historic highs, "smart money" starts selling rather than adding positions. Because the framework tells them mean reversion is inevitable.

Western Digital has surged more than 200% this year and over 600% in the past 12 months. Given this gain, "it's risen too much" alone justifies selling. No bad news is needed—only the expectation that "the good news is already priced in." But here's a key question.

If HDD demand is no longer "bell-shaped" but becoming a monotonically increasing curve, the "cyclical" framework fails. And if the framework fails, "mean reversion" no longer applies, meaning the assumption that "a peak must be followed by a trough" no longer holds. Irving Tan said something on the earnings call that might be the most important line of the entire event.

He said, "Computing cycles can be reused, but data grows compoundingly." The implications of this statement are worth exploring. GPU demand is "cyclical." After training a model, computing power can be reused for the next model. Demand has a ceiling, with peaks and troughs. But HDD demand is "compounding." Every inference, every agent workflow, and every autonomous driving log generates new data. This data doesn't disappear—it only accumulates. Training generates initial datasets, inference generates ongoing logs, agents generate multi-step workflow data, and physical AI even requires "manufacturing" synthetic data.

Data isn't "consumed"—it's "accumulated."

Consumption has cycles; accumulation does not. If this logic holds, the HDD demand curve isn't "bell-shaped" but "exponential." Using a cyclical framework to understand an exponential curve naturally leads to the conclusion that "it's risen too much." Because the cyclical framework assumes the curve has a peak, while an exponential curve does not.

Of course, there's a fair rebuttal. "Data compounding" is a long-term logic, while stock prices are short-term priced. Even if long-term demand grows exponentially, short-term supply tightness, price elasticity, and customer inventory cycles can still cause volatility. The 2022–2023 downturn wasn't an illusion—it really happened. But the key difference is that the 2022–2023 downturn was driven by "shrinking PC and consumer electronics demand." The current growth is driven by "structural demand from AI data centers." The demand drivers are different, with different elasticities and cyclicality. Storage purchases by hyperscale data centers don't fluctuate as much with economic cycles as consumer electronics do.

The essence of the problem is that the market is using a yardstick designed for an "old demand structure" to measure a company under a "new demand structure."

03

If "data compounding" is still theoretical, the next three pieces of evidence are more concrete.

The first piece of evidence is long-term agreements (LTAs).

Irving Tan revealed on the earnings call that Western Digital is discussing LTAs with customers for 2029, 2030, and even 2031. He said, "Customer demand for extending LTAs to 2031 remains very strong."

The traditional HDD business model is "spot trading." Customers purchase as needed, prices fluctuate with supply and demand, and gross margins swing wildly. This is a classic "commodity" model. LTAs change everything. When customers lock in capacity demand for the next three to five years, HDDs shift from "spot" to "long-term contracts," from "commodities" to a form of "subscription."

This change has fundamental implications for valuation. Cyclical stocks typically trade at 8 to 15 times earnings because profits are unpredictable. Utility stocks typically trade at 20 to 25 times earnings because revenue and cash flow are predictable. Power companies enjoy stable valuations because they sign long-term power purchase agreements, constraining revenue. If HDD companies can also sign five-year LTAs, their revenue predictability approaches that of power companies.

Of course, LTAs carry risks. What's the pricing mechanism? Will customers renege if future demand falls short? Does long-term contracting cede some pricing power? These questions haven't been fully tested yet. But the direction is clear. When HDDs sign five-year contracts, they're no longer traditional "cyclical stocks."

The second piece of evidence is no capacity expansion.

Demand grew 44% year-over-year—why not expand capacity?

Because not expanding is the optimal way to maximize profits. Expanding capacity means increasing supply, balancing supply and demand, lowering prices, and reducing gross margins. Not expanding, combined with technological iteration, means supply remains constant, supply-demand tightness persists, prices stay elevated or rise, and gross margins continue to improve.

Western Digital's technology roadmap is clear. The 40TB UltraSMR entered volume production in the second half of 2026. HAMR drives are expected to complete customer certification and enter volume production in 2027, with first-generation capacities exceeding 40TB. By 2029, the company aims to achieve 100TB per drive in volume production. More capacity in the same physical space reduces cost per TB, but price per TB doesn't drop—it rises. Volume remains constant, prices rise, and costs fall. Profits grow threefold. This quarter's capital expenditures were only $108 million, accounting for less than 3% of revenue. This ratio is nearly the lowest in the semiconductor and storage industry. It means Western Digital doesn't need massive new capacity to meet demand growth—technological iteration itself is "capacity expansion." More critical is the industry structure. Only three HDD manufacturers remain globally: Seagate, Western Digital, and Toshiba. After the brutal 2022–2023 downturn, all three have become extremely "capacity-disciplined." None is willing to expand aggressively first. This tacit understanding (tacit agreement) is the strongest moat in an oligopoly. Not expanding isn't "conservatism"—it's "maximizing pricing power." In an oligopoly, restraint is the moat.

The third piece of evidence is physical AI.

The earnings call included an easily overlooked incremental detail. Irving Tan mentioned that an autonomous driving company's exabyte-scale storage demand will "grow several times" in the 2026 calendar year. Physical AI differs from generative AI in a key way. Generative AI's training data comes from the internet—text, images, videos. While large, this dataset is finite. Physical AI's training data comes from the real world—sensors, cameras, radar. Real-world data is vastly insufficient. The solution is using AI to generate synthetic data. Data isn't "collected"—it's "manufactured." "Manufacturing data" means data growth has no physical upper limit. Generative AI is HDDs' first growth curve; physical AI is the second. The baton has been passed. LTAs lock in "volume," no capacity expansion locks in "price," and physical AI locks in "the future." Together, these three locks mean HDDs are no longer traditional "cyclical" assets.

04

After understanding why the "cyclical" framework is failing, the next question is when the market will switch to a new yardstick. Framework shifts don't happen overnight. They require time, continuous validation, and a "trigger event."

First, let's examine Western Digital's "three acts" in the capital markets.

Act 1, 2015–2023: Abandoned. Under the narrative that SSDs would crush HDDs, Western Digital was stamped with a "sunset industry" label. It traded at 8 to 12 times earnings, similar to utilities. The market priced it as "a legacy technology being replaced."

Act 2, 2024–early 2026: Rediscovered. AI data centers' demand for high-capacity storage exploded, transforming HDDs from "sunset" to "scarce." The stock price surged from $50 to over $600. The market priced it as "a core beneficiary of AI infrastructure."

Act 3, August 2026: Questioned again. Despite beating expectations across the board, the stock dropped 11% after the market closed. The market priced it as "it's risen too much; it must have peaked."

The cruelty of the word "again" is that the market has never truly believed in Western Digital. The first time, it didn't believe because "HDDs are the past." The second time, it didn't believe because "it's risen too much."

The reasons changed, but the conclusion didn't. But there's a historical precedent worth noting. Before the 1990s, the power industry was also seen as "cyclical." When the economy did well, electricity demand rose; when it didn't, demand fell. After long-term power purchase agreements became widespread, power companies transformed into "utilities" with predictable revenue and stable valuations. HDD LTAs are the storage industry's equivalent of PPAs.

Kikkoman also went through a "domestic market peaked, valuation compressed" phase in the 1970s. But through globalization and category expansion, it completed a shift from "cyclical" to "compounding." Western Digital's "globalization" is the global expansion of AI data centers; its "category expansion" is from cold storage to physical AI.

What conditions are needed for a framework shift?

First, consecutive quarters of validation. LTAs continue to materialize, and gross margins keep rising.

Second, at least one "cyclical downturn" test. If Western Digital doesn't fall during the next demand slowdown, the market will truly believe it's "non-cyclical."

Third, sell-side analysts' "framework upgrade." A shift from "cyclical" to "platform" rating logic. We lean toward believing this "yardstick switch" could take two to four quarters. Until then, every "beat" might be punished by "profit-taking." This isn't a fundamentals issue—it's a cognitive update speed issue. Of course, there's a fair risk to mention.

The long-term threat of substitution for SSDs is real. If the cost of NAND flash memory continues to decline, at a certain tipping point—such as when the price gap per TB narrows to less than double—HDD's "cold data" market could be eroded. The oligopoly's tacit understanding is also fragile. If one player decides to aggressively expand production, the balance will be broken. These risks alter the slope, not the direction.

05

Profits soared 12-fold, only to plunge 11% after hours. If you measure it with a "cyclical stock" ruler, it indeed looks "unappealing." Gross margins surged from 25% to 54%, and the stock price has risen more than 600%, with expectations fully priced in. Every peak in a cycle is followed by a trough. The 11% after-hours drop reflects "smart money" cashing out.

If you measure it with a "data compounding" ruler, the story is just beginning. Data only grows, LTA is locked in until 2031, the second curve of physical AI is just launching, and there is still room for gross margins to improve. Not expanding production means pricing power is only strengthening. This is not a company "at the peak of a cycle" but one whose "business model is shifting from cyclical to compounding."

Two rulers, two conclusions. The market is falling because most people are still using the first ruler.

Western Digital delivered a earnings report with profits surging 12-fold. The market responded with an 11% plunge. This is not a problem with Western Digital but with the ruler being used. The data says things have never been better. The stock price says, "You've risen too much." The truth lies somewhere in between. It's not that the company has worsened; it's that the market isn't ready to accept a "non-cyclical" Western Digital yet.

The process of "acceptance" is precisely where the largest expectation gap lies. The question is, when will the market switch rulers? When the news of LTA signed until 2031 is fully digested by more analysts, when the EB-scale demand for physical AI materializes in FY2027, and when gross margins stay above 55% for four consecutive quarters, the "cyclical" label will fall off on its own. But until then, every "beat" could be met with a "plunge." That is the cost of switching rulers—and the source of the expectation gap.