Produced by I Xia Hai Fallsea Written by I Hu Buzhi

After market close on August 5 (ET), SanDisk (NASDAQ: SNDK) released its Q4 FY2026 financial results. Revenue reached $8.97 billion, up 372% YoY and 51% QoQ, surpassing market expectations of $8.39 billion. GAAP net profit was $6.9 billion, compared to a loss of $23 million in the same period last year. Adjusted EPS was $39.25, higher than analyst expectations of $34.37. The board also approved a $14 billion share repurchase plan on the same day.

The results exceeded expectations across the board. However, the most noteworthy aspect of this report was not revenue, growth rates, or the $14 billion buyback.

It was a single number.

84.6%.

GAAP gross margin stood at 84.6%, up 6.2 percentage points QoQ and 58.4 percentage points YoY.

Let's zoom out. Over the past two decades, NAND flash memory gross margins typically fluctuated between 20%-50%. During the 2019 downturn, they hovered around 20%-30%. In the 2022 upturn, they reached 40%-50%. SanDisk's own historical peak never exceeded 55%. 84.6% is not a "cyclical high."

This is a number never seen before in the NAND industry. It deviates from the historical average by over 40 percentage points. A company that started by selling USB drives and SD cards now has a higher gross margin than NVIDIA.

A never-before-seen number implies either a "paradigm shift" or a "final frenzy." The core question is whether SanDisk's 84.6% represents NVIDIA-style "ecosystem lock-in" or temporary "premium pricing" due to supply-demand mismatches.

The former signifies a strategic upgrade; the latter, a cyclical illusion. We lean toward the latter, at least for the next 2-3 years.

01

Let's first examine how 84.6% came about.

Revenue grew 51% QoQ. Management broke down this growth during the earnings call: roughly one-third came from increased shipments, and two-thirds from price hikes. In other words, SanDisk's additional earnings this quarter came primarily from "higher prices" rather than "higher volumes."

This structure is critical. It means the 84.6% gross margin resulted not from "economies of scale" but from "pricing power." More precisely, pricing power derived from a "supply-demand gap."

What happened on the demand side? Explosive demand for enterprise SSDs from AI servers. This quarter, data center revenue reached $2.98 billion, up 1,298% YoY. Edge computing revenue hit $5.43 billion, up 392% YoY. CEO David Goeckeler stated on the call, "The expansion of AI workloads is driving accelerated growth in enterprise SSD demand across the ecosystem, particularly for inference, which significantly increases NAND content per deployment."

What happened on the supply side? Samsung and SK Hynix converted significant NAND production lines to HBM manufacturing. HBM offers higher profit margins, making this a rational business decision. However, its side effect was a contraction in effective NAND supply. Meanwhile, NAND capacity expansion takes 2-3 years, preventing rapid short-term responses.

After a brutal destocking phase in 2023, companies became extremely "capacity-disciplined."

Demand surged, supply contracted, and capacity expansion lagged. Combined, these factors caused NAND wafer prices to skyrocket. SanDisk's contract prices rose 50% in a single November 2025 increase, with even more aggressive hikes in the first half of 2026. Thus, the cause of 84.6% is clear: it's not that SanDisk "got stronger," but that NAND "became scarce." However, "scarcity" and "strength" are not the same.

Scarcity is market-driven; strength is self-built. What the market gives, it can take away.

A comparison is needed here. NVIDIA's gross margin hovers around 75%. TSMC's around 60%. SanDisk's 84.6% exceeds both. But NVIDIA's 75% reflects a decade-long CUDA ecosystem lock-in effect, while TSMC's 60% stems from advanced node technological lead and $20 billion-per-fab capital barriers.

What lies behind SanDisk's 84.6%?

A supply-demand gap. This is where the question begins.

02

The market naturally compares SanDisk's gross margin surge to an "NVIDIA moment"—a strategic upgrade from "commodity" to "AI infrastructure core." This analogy has merit but contains a fatal flaw.

First, let's examine the similarities.

Before 2022, NVIDIA's gross margin hovered around 60%-65%, identifying as a "gaming GPU company." From 2023-2024, AI compute demand exploded, boosting margins to 75% and transforming its identity to "AI compute monopolist."

SanDisk's gross margin hovered around 30%-40% before 2024, identifying as a "USB drive and SD card company." From 2025-2026, AI storage demand surged, lifting margins to 84.6% and transforming its identity to "AI data center storage supplier." Both completed "new identity" upgrades amid the AI wave, with significant margin surges driven by AI infrastructure expansion. The analogy holds superficially.

However, it only describes superficial similarities without addressing underlying structural differences.

Where do these differences lie?

First, the source of pricing power differs. NVIDIA's 75% margin rests on its CUDA ecosystem, with millions of developers globally training on CUDA, all major AI frameworks prioritizing CUDA compatibility, and client software stacks built around CUDA. Migration costs are prohibitively high. Clients don't "choose not to switch"—they "can't afford to switch." SanDisk's 84.6% rests on a supply-demand gap. Clients buy SanDisk's NAND not because "only SanDisk can make it" but because "it's the only stock available now." Once Samsung, SK Hynix, and Micron release capacity, clients can switch immediately. NAND is a standardized product. Samsung's NAND and SanDisk's NAND are nearly indistinguishable to clients.

Second, competitive landscapes differ. NVIDIA holds over 80% market share in AI GPUs, approaching monopoly status. SanDisk holds about 15%-20% in NAND, with Samsung at ~35%, SK Hynix at ~20%, Micron at ~25%, and Kioxia also a major player. This represents an oligopoly with 4-5 competitors, not a monopoly. If Samsung expands NAND capacity, can SanDisk sustain 84.6%? Probably not.

Third, technological barriers differ. A clear performance gap exists between NVIDIA's Blackwell architecture and AMD's MI300. Clients choose NVIDIA because "only NVIDIA delivers this performance." No such qualitative gap exists in NAND flash memory. While 3D NAND layer counts race upward (232, 300, 400 layers), these represent quantitative, not qualitative, differences. Clients won't become "SanDisk-exclusive" just because its NAND has 20 more layers.

Thus, a more accurate analogy isn't an "NVIDIA moment" but "TSMC circa 2020." Around 2020, TSMC's gross margin rose from 50% to 60%, driven by tight advanced node capacity with Apple, NVIDIA, and AMD vying for production slots. However, TSMC's 60% was sustainable due to three pillars: technological lead (only TSMC could make 3nm chips), client lock-in (high switching costs after chip design tape-out), and capital barriers (over $20 billion per advanced fab).

SanDisk's 84.6% lacks any of these three conditions. NAND has no "only I can make this" technological lead. Clients can switch suppliers anytime. NAND fab investment intensity is far lower than advanced nodes.

NVIDIA's 75% is an "ecosystem tax." As long as CUDA exists, so does 75%. SanDisk's 84.6% is a "scarcity tax." As long as the supply-demand gap exists, so does 84.6%. But supply-demand gaps are temporary.

03

When does this "temporariness" end?

To answer, we must examine supply-side responses.

An 84.6% gross margin signals a clear invitation to the industry: expand capacity. When NAND margins surge from 40% to 84.6%, what will Samsung, SK Hynix, and Micron do? Capital instincts take over.

However, NAND capacity expansion involves "time lags." New fabs take 2-3 years. Converting existing lines from HBM back to NAND takes 6-12 months. Technological upgrades from 232 to 300+ layers take 1-2 years. This means even if all players hit "expand" simultaneously today, new supply would only concentrate release in H2 2027-2028.

But "time lags" don't mean "won't happen"—they merely delay arrival.

More notably, SanDisk's own guidance signals the first crack.

Next quarter's guidance: revenue $10.3-$10.8 billion (midpoint $10.55 billion) vs. market expectations of $11.16 billion—a gap of ~$600 million (~5.5%). EPS guidance: $44-$46 vs. expectations of $45.58 (basically in line). Gross margin guidance: 83%-85%, roughly flat or slightly down from Q4's 84.6%.

Revenue guidance missed expectations. The 5% gap falls within normal volatility, but it signals slowing growth momentum. This quarter's revenue grew 51% QoQ; next quarter's midpoint implies ~17.6% growth. The deceleration from 51% to 17.6% is pronounced.

Of course, seasonal factors and client purchasing rhythm adjustments play a role. We can't conclude a "turning point" from one quarter's guidance alone. But the market is already pricing in this possibility.

In July, SanDisk's stock plummeted 47% in a single month. SK Hynix fell 54%, Samsung Electronics 42%, and Micron 33%. The entire memory sector underwent collective valuation compression in one month. This isn't an isolated company issue—it reflects loosening market pricing of the "memory supercycle" thesis.

After market close on August 5, SanDisk initially fell 8% but reversed to close up over 5% the next day. This "drop-then-rally" pattern itself indicates massive market divergence: some are "taking profits," others "buying the dip."

Let's conduct a scenario analysis. Under an optimistic scenario, AI demand continuously exceeds expectations, supply-side capacity discipline holds, and gross margins decline gradually from 84.6% to 80%, then 75%. This scenario has ~20% probability. Under a base-case scenario, AI demand remains high but supply-side capacity releases in H2 2027, pulling margins down to 65%, then 55% within two years. This has ~55% probability. Under a pessimistic scenario, AI capex slows combined with aggressive supply-side expansion, crashing margins to 45% within 18 months. This has ~25% probability.

Under the base-case scenario, SanDisk's gross margin will return to the 60%-65% range by FY2028—still far above NAND's historical 35% average but no longer 84.6%.

84.6% is not "NAND's new normal"—it's "NAND's supply-demand peak." Peaks only go downward. The question isn't "will it decline" but "when will it decline" and "how far will it fall."

04

However, we must fairly discuss a counterpoint.

SanDisk disclosed a key detail on the earnings call: it has signed NBM (New Business Model) long-term agreements with eight data center and edge computing clients. NBM represents multi-year fixed-price supply contracts, with weighted average terms exceeding four years (up to five years). These agreements cover over 50% of FY2027 supply and ~66% of FY2028 supply, using a fixed-price plus floating pricing mechanism. Management stated guaranteed revenue totals $93.9 billion. CEO Goeckeler said in Q3, "We're turning this business into recurring revenue." Q4 provided concrete contractual support for this claim.

This information requires careful consideration.

If NBM agreements fully materialize, SanDisk's business model shifts from "spot trading" to "long-term supply contracts."

Spot trading represents a classic "commodity" model, with prices fluctuating with supply-demand and gross margins volatile. Long-term contracts resemble "subscription" or "utility" models, with predictable revenue and cash flows. This aligns with the "LTA" logic we discussed when analyzing Western Digital. When HDDs sign five-year LTAs, they cease being traditional "cyclical stocks." Similarly, when NAND signs four-to-five-year contracts, its "cyclicality" weakens. However, NBM and NVIDIA's CUDA represent two entirely different types of "lock-in."

CUDA locks in the "software ecosystem," with migration costs involving code rewrites. NBM locks in "supply contracts," with breach of contract costs involving legal penalties. The former's stickiness is technological; the latter's, contractual. Contracts expire, can be breached, or renegotiated under extreme market conditions.

Technological ecosystems, once established, are nearly irreversible. Thus, while NBM is a positive signal, it modifies "cycle amplitude" rather than "cycle nature." It flattens margin declines but doesn't reverse their direction. Another structural issue remains.

SanDisk's consumer business revenue was $556 million this quarter, down 5% YoY and 32% QoQ. Consumer products now account for ~6% of total revenue. USB drives, SD cards, and consumer SSDs—these "old SanDisk" offerings are shrinking. This confirms SanDisk's "old identity" is fading. It's no longer a "USB drive company."

Data centers and edge computing now contribute 94% of revenue. This is an AI infrastructure company. However, "identity upgrades" require not just "revenue mix shifts" but "moat construction." Currently, SanDisk's moat remains "supply-demand gap" plus "NBM contracts," not "technological monopoly" or "ecosystem lock-in." The journey from "supplier" to "platform" still has miles to go.

05

84.6%. It's a number never seen before in the NAND industry. It let a "USB drive company" surpass NVIDIA in gross margin. It propelled SanDisk from a 52-week low of $40 to over $2,300 in one year. It made a company spun off just 18 months ago approach a $180 billion market cap.

However, the phrase 'never seen before' carries two meanings. One is 'an unprecedented new height,' and the other is 'an irreplicable peak.' We tend to believe that 84.6% is closer to the latter. Not because it is 'bad.' Of course, 84.6% is good, genuine, and direct evidence of the explosion in AI storage demand. A company earning a net profit of $6.9 billion in a single quarter deserves respect no matter what. Rather, it is because it is 'unsustainable.' Its support comes not from 'ecosystem lock-in' but from a 'gap between supply and demand.' This gap will be closed.

On the day the gap is closed, the gross margin will decline. But a decline does not equal a collapse. Returning from 84.6% to a range of 60% to 65% would still represent an excellent level never before seen in the NAND industry. SanDisk's 'identity shift' is real. It has indeed transformed from a 'commodity' to an 'AI infrastructure.' However, 84.6% is not the steady state of this 'new identity' but rather the peak during the 'identity transition period.'

Just as Tesla's gross margin peak in 2021 was not the 'norm for electric vehicles' but rather a peak driven by 'supply-demand mismatch plus subsidies,' its subsequent decline to a range of 18% to 20% represents the steady state. As for where SanDisk's steady state lies, we tend to believe it is in the range of 60% to 65%.

The market often prices a 'return to normal' as if it were a 'collapse.' This is the 'curse' of 84.6%. The higher it is, the more the future decline resembles a crash. Not because the company has deteriorated, but because expectations have been set too high. The 47% plunge in July was not fundamentally a punishment for 'performance.' Performance had not yet been announced at that time.

It was a punishment driven by fear of a 'cyclical peak.' The market was asking, 'What comes after 84.6%?' The earnings report on August 5 provided an answer. After 84.6% comes a range of 83% to 85%. At least for the next quarter, the gross margin can be maintained. But what about after that? What about FY2028? What about FY2029?

When new capacity from Samsung and SK Hynix is released, when the squeeze on NAND production lines by HBM eases, and when the fixed-price portions of NBM contracts begin to fall below market spot prices, the mean reversion of gross margins will begin. Is 84.6% NAND's 'Nvidia moment'? Not entirely.

The essence of an 'Nvidia moment' is ecosystem lock-in, and SanDisk does not yet have its own 'CUDA.' Is 84.6% the 'final hurrah'? Not entirely either. A 'final hurrah' implies nothing comes after. But for SanDisk, a range of 60% to 65% still awaits. These would still represent the best days ever seen in the NAND industry. The truth lies somewhere in between. 84.6% is the peak of the identity shift, not the starting point nor the endpoint.

The question is, when will the market distinguish between the difference between a 'peak' and an 'endpoint.' Until then, every decline in gross margin will be punished as if it were a collapse. Every marginal slowing in guidance will be interpreted as the end of the cycle. This is the price of 84.6%. It is also the source of expectation gaps.