The Ledger of NAND: Jane Street's Position and the AI Storage Cycle

On a recent 13F filing, a quantitative trading giant disclosed a 540% increase in its stake in a newly independent NAND flash manufacturer. The position now stands at approximately 7.41 million shares. Over the past twelve months, the equity has appreciated 3,000% before retracing 36% from its apex. The market has spoken; the ledger does not lie, only the interpreters do. This company, SanDisk, was carved out of Western Digital's flash business in 2025. The market is treating it as a pure-play AI infrastructure supplier. The numbers are impressive. Data center revenue reached 38% of total revenue in fiscal 2026, up 437% year-over-year. The company has secured long-term supply agreements worth $93.9 billion with eight customers, including three major U.S. cloud providers. On the surface, this is a straightforward AI storage thesis. But my due diligence requires dissecting the structural shifts beneath this headline growth. The core insight, based on my experience modeling liquidity risks across markets since 2017, is that SanDisk represents a case study in the institutional integration of hardware into the AI capex cycle. The demand profile is changing. AI inference, not just training, is becoming the primary driver of NAND consumption. This is a different beast than the smartphone cycles of the last decade. It requires lower latency, higher bandwidth, and larger capacities. The strategic pivot is evident in their development of High Bandwidth Flash, a new packaging technology for AI inference workloads, positioned as a potential differentiator against competitors like Samsung and SK Hynix. The sector's financial architecture shows it is operating in a healthy cycle. NAND contract prices rose 50-60% in 2024 and are projected to rise another 10-20% in 2025. Channel inventory sits at 4-6 weeks, down from 8-10 weeks at the 2023 highs. The industry is in a re-stocking phase. The technology gap with the market leaders is real but finite. The SanDisk-Kioxia alliance is roughly 6-12 months behind in 3D NAND layer counts. This is a tactical disadvantage, not a strategic one. Their HBF roadmap could narrow the gap in specific AI workloads, but the capital expenditure intensity remains high. New factory depreciation will pressure gross margins by 3-5 percentage points until utilization reaches 70-75%. The contrarian angle is not whether AI will demand storage—it will. The question is the quality of the revenue. A significant portion of that $93 billion backlog is with three hyperscale customers. This concentration is a double-edged sword. It provides near-term visibility and secures demand, but it is a direct contract with the same companies whose capital expenditure is the cyclical risk. The leading-edge memory makers are also investing heavily, and history suggests they will not yield the AI storage market without a fight. Rebalancing is not panic; it is preservation. A deeper concern is the current valuation. The market is pricing in a smooth execution of this AI transition with an EV/EBITDA multiple of 15-20x. This reflects a high level of confidence. It ignores that a substantial portion of the demand is tied to an AI capex cycle that can turn. The long-term supply agreements provide a floor but they also limit the upside from the very spot market volatility that analysts often favor. Every bull run is a tax on due diligence. The signal from the market is loud, but so is the noise. The structure of this cycle is built on a different foundation than those of 2018 or 2022. The demand is real. The shift from consumer to enterprise is a structural change. Yet the industry's history is a story of boom and bust. The question is not whether SanDisk will be a key player in the AI storage buildout, but whether the current valuation already accounts for every possible outcome. It is the difference between the price and the value. I would watch the next two quarters for the AI inference market. If the HBF samples appear on time and the cloud providers continue to expand storage budgets, the thesis will hold. But if the capex slows or a competitor closes the technology gap, the current margin of safety will evaporate. The ledger is not yet balanced.