Hook: The 84.6% Anomaly
Most people in the semiconductor analyst community look at SanDisk's 84.6% gross margin and immediately attribute it to technical superiority. A 62-percentage-point surge in five quarters—from 22.5% to 84.6%—seems like the smoking gun of a dominant product portfolio. But I've spent enough time auditing smart contract architectures to know that extreme metrics often hide structural vulnerabilities rather than strengths. The question isn't whether SanDisk has good technology; it's whether the market is mispricing the underlying risk. A 84.6% margin in NAND, an industry that has historically swung between 10% and 60% in a cycle, is not a new normal. It's a signal that the supply-demand equilibrium has broken in a way that may not be sustainable.
Context: The AI Storage Land Grab
SanDisk, spun off from Western Digital in early 2025, is a pure-play NAND Flash IDM with a joint development agreement (JDA) with Kioxia. The company's core product is the BiCS series of 3D NAND, currently at BiCS8/BiCS9 generation with 218–300+ layers. The AI data center boom—training clusters needing 8–20TB NVMe SSDs per node, and the inference explosion requiring even more capacity per server—has created a demand shock that the NAND industry was not prepared for. After the 2023 market collapse, where NAND revenues dropped 40% and all major players bled cash, the industry slashed capital expenditures. By 2025, when AI demand hit, supply was structurally constrained. SanDisk reported that two-thirds of its revenue growth came from price increases, not volume. That's a classic sign of a supply-constrained market. The 84.6% margin is a snapshot of an extreme point in the cycle, not a reflection of enduring competitive advantage.
Core: Decomposing the Margin—Supply Scarcity Over Technology
Let's get granular. SanDisk's gross margin trajectory from 22.5% in Q1 FY2025 to 84.6% in Q2 FY2026 is the steepest recovery in NAND history. But the company's own guidance is for 80% gross margins going forward—a 4.6-percentage-point haircut that management frames as conservative. From my experience analyzing DeFi protocol tokenomics, such a forward guidance "sacrifice" is often a hedge against two hidden forces: depreciation drag from new capacity, and the inevitable price normalization. SanDisk's customer contracts—eight multi-year agreements covering 50% of FY2027 shipments and 66% of FY2028—are a double-edged sword. They lock in volume and price floors, but they also limit upside in a spot market that could remain hot. More importantly, the contracts reveal that SanDisk's customers are terrified of future shortages. This is a market where the buyer is locking in supply, not the seller locking in price. That imbalance is a classic top-of-cycle indicator.
Contrary to the "technical leadership" narrative, the margin expansion is overwhelmingly driven by the collapse of NAND supply in 2023–2024 and the subsequent demand surge. The actual technology gap between SanDisk/Kioxia and Samsung/SK Hynix is approximately 0.5–1 generation (6–12 months). That's not a wide enough moat to justify a 84.6% margin when history shows that NAND margins above 60% are rare and fleeting. The 2026 margin is a product of the specific intersection of AI demand, industry supply discipline, and the lag in new capacity from YMTC (which is building a third fab that could add ~10% of global NAND capacity by 2027). If YMTC's capacity comes online faster than expected, or if AI demand growth decelerates, the margin compression will be brutal.
Core: The Missing Puzzle Piece—HBM Absence and Depreciation
SanDisk's technical portfolio has a glaring gap: it does not produce HBM (High Bandwidth Memory). Samsung and SK Hynix offer full-stack AI storage solutions combining HBM for GPU acceleration and NAND for data persistence. This composability of memory and storage is becoming a key criterion for hyperscalers. SanDisk's reliance on third-party HBM (from competitors) creates a structural disadvantage in AI ecosystem lock-in. The company's "leading technology portfolio" claim is defensible in NAND packaging and firmware, but not in the broader AI memory hierarchy.
More importantly, the current 84.6% margin is artificially inflated by low depreciation. SanDisk's existing fab equipment is largely depreciated (5–7 year straight-line). Any new capacity investment will bring fresh depreciation that drags gross margins by 5–10 percentage points. The 80% guidance is likely already factoring in this depreciation over the next 18 months. If the company ramps new fabs to meet the 2028 contracted volumes, depreciation could push actual margins below 70% even in a strong price environment. The recent quarterly results show that revenue grew 51% sequentially, but that was driven 2/3 by price, only 1/3 by volume. Volume growth is constrained by capacity. The company is not investing aggressively in new capacity—a signal that management is prioritizing free cash flow and share buybacks over capturing market share. This is rational but risky: it leaves the door open for YMTC and others to fill the gap.
Contrarian: The Blind Spots Everyone Ignores
First, the customer concentration risk. The eight contracts cover 50% of FY2027, but that means the other half is exposed to spot market volatility. If the market turns, SanDisk has no downside protection for that portion. Second, the Kioxia JDA dependency. SanDisk's technology roadmap is co-developed with Kioxia; any renegotiation of the JDA (e.g., if Kioxia prioritizes its own brand SSD) could cut off the technology pipeline. This is a single-point-of-failure that IDM competitors like Samsung do not have. Third, the YMTC Xtacking threat. While YMTC faces equipment sanctions, its technology (wafer bonding for higher density) is innovative. If YMTC achieves volume production by 2027 despite sanctions, the global NAND supply could increase by 10%+ in a market that is just beginning to normalize. The impact on pricing would be immediate.
Composability isn't just a technical feature; it's a ecosystem. In NAND, the ecosystem is the ability to offer a complete AI storage solution—HBM + NAND + controller + firmware. SanDisk is missing HBM. That is a strategic blind spot that will become more obvious as hyperscalers demand tighter integration. The company's current success is a bull market phenomenon riding on AI demand, but the underlying architecture is not as future-proof as the narrative suggests.
Takeaway: The 2027–2028 Inflection Point
SanDisk's 84.6% margin is a snapshot of a cyclical peak, not a structural shift. The company's ability to lock in long-term contracts buys time, but the fundamental dynamics of NAND—commodity-like pricing, binary supply-demand swings, and the looming threat of YMTC capacity—remain unchanged. The real test will come in FY2028, when the contracted volumes reach two-thirds of shipments and new capacity begins to hit the market. If AI demand growth slows by even 10%, the margin will compress faster than the guidance implies. The narrative today is built on a scarcity premium. We don't need more NAND; we need more visibility into the supply-demand balance. Until then, treat the 84.6% as a mirage—beautiful, but dangerous to trust.