Let's be clear: a 540% position increase in a freshly spun-off NAND manufacturer is not a conviction play. It's a signal. The question is — signal for what?
Jane Street's Q3 13F filing shows the quantitative trading giant accumulating SanDisk (SNDK) at a pace that dwarfs its typical semiconductor exposure. This isn't a retail momentum chase. This is a firm that models latency, liquidity, and information asymmetry deciding that the risk/reward on this specific storage play has shifted materially.
Here is the data: SanDisk, spun off from Western Digital in February 2025, is now a pure-play NAND flash manufacturer. The company has locked in $93.9 billion in long-term supply agreements with eight customers, including three major US cloud providers. Data center revenue grew 437% year-over-year in fiscal 2026. And they're developing a product category — High Bandwidth Flash (HBF) — that directly targets the AI inference bottleneck that HBM can't solve.
Most retail traders see a memory chip stock trading at 30x earnings and think "bubble." I see a company that has transformed its revenue profile from cyclical spot-market exposure to contracted, utility-like cash flows. That's not a narrative. That's a structural change in how the market should price this asset.
But here's where it gets interesting: the market hasn't fully repriced this yet. The PE ratio is elevated, sure. But the earnings multiple is based on trailing earnings that don't yet reflect the full ramp of those long-term contracts. The market is still valuing SanDisk like a memory company. The contracts suggest it should be valued like an infrastructure play.
That's the arbitrage. And Jane Street is not the only one seeing it.
The Context: A Spinoff at the Exact Moment AI Storage Demand Inflects
SanDisk's history is a masterclass in corporate timing. Western Digital announced the separation in 2023, executed it in February 2025, and the newly independent company emerged into the strongest NAND pricing environment in a decade.
Here's the context you need:
- The AI inference problem: Training models is a one-time cost. Running them continuously is a recurring cost. Every ChatGPT query, every AI-powered search, every autonomous vehicle decision requires model parameters to be loaded from storage into compute. That's a NAND problem.
- The HBM bottleneck: High Bandwidth Memory is expensive, power-hungry, and supply-constrained. It's optimized for training. Inference needs a different profile — high capacity, moderate bandwidth, low cost per bit. That's exactly the gap HBF targets.
- The supply discipline: The NAND industry has learned from the 2022-2023 crash. Samsung, SK Hynix, and Micron have all signaled capex discipline. SanDisk's capacity utilization is running at 85-95%, and the industry is in the early stages of a restocking cycle.
SanDisk's positioning is unique: it's the only pure-play NAND company with meaningful scale. Samsung has foundry and logic. SK Hynix has HBM and DRAM. Micron has DRAM. SanDisk is NAND, and only NAND. When AI inference demand drives NAND prices up 20-40% (which is my base case for 2026), SanDisk's earnings leverage is unmatched.
The Core: What the 13F Filing Actually Reveals About Order Flow
Let's get into the mechanics. Jane Street's 540% increase in SNDK holdings isn't just a directional bet. It's a liquidity provision strategy.
Here's what I know from my own experience trading around 13F filings: quant firms like Jane Street don't accumulate positions the way a mutual fund does. They build inventory to facilitate client flow. When a stock is newly spun off, there's a period of dislocation — index funds are forced to sell, active managers are underweight, and the options market hasn't fully calibrated.
That's the window Jane Street operates in. They're not betting on NAND prices. They're betting on the spread between where the stock trades and where it should trade once the market fully digests the spinoff dynamics.
But here's the part most analysts miss: the $93.9 billion in long-term contracts changes the fundamental character of the business. Let me break this down:
Revenue visibility: Eight customers, three of which are hyperscale cloud providers, have committed to multi-year purchase agreements. This isn't a spot-market business anymore. The revenue is contracted. The pricing is largely locked. The volume is committed.
The 437% data center growth number: This is the headline that matters. It tells you that the AI inference demand is not theoretical — it's showing up in actual purchase orders. Cloud providers are not signing $90+ billion contracts based on speculative demand. They're signing them because they have capacity constraints on AI inference storage today.
The HBF optionality: SanDisk plans to sample High Bandwidth Flash to customers next year. If this product works — and I've spent enough time analyzing storage architectures to believe the technical approach is sound — it opens an entirely new market that doesn't exist today. HBM is a $20+ billion market. HBF could be comparable, and SanDisk is the only player with a credible product in development.
Now, let me give you the technical breakdown that most coverage misses:
The 218-layer BiCS8 architecture: SanDisk's current generation uses 218-layer 3D NAND, produced in partnership with Kioxia at their Flash Ventures fabs. This puts them roughly 0.5-1 generation behind Samsung and SK Hynix in layer count. But here's the thing: layer count is a lagging indicator. The leading indicator is bit density and cost per bit, and SanDisk's QLC (4-bit per cell) technology is industry-leading.
The CBA (CMOS Bonded Array) architecture: This is the next-generation approach that separates the CMOS logic from the memory array and bonds them together. It's the same fundamental approach that enables HBM's high bandwidth, applied to NAND. SanDisk and Kioxia have been developing this for years, and it's the foundation for both their 300+ layer roadmap and HBF.
The DUV advantage: Unlike logic chips that require EUV lithography, NAND manufacturing uses mature DUV immersion tools. This means SanDisk is largely insulated from the export control regime that's constraining advanced logic and AI chip manufacturing. Their supply chain is secure, and their capex per wafer is significantly lower than leading-edge logic.
The depreciation curve: Storage equipment is depreciated over 5-7 years. With capacity utilization above 80%, depreciation is fully covered. As the 218-layer products ramp and yields improve, the operating leverage is substantial. My estimate is that gross margins expand from the current 25-35% range to 30-40% by 2026.
The Contrarian Angle: Why the Market Is Wrong About SanDisk's Valuation
Here's where I diverge from the consensus. The market is treating SanDisk's 25-35x PE as a sign of froth. I think that's a misread of the business model transformation.
Let me walk through the logic:
The "utility" argument: A company with $93.9 billion in contracted revenue, locked-in pricing, and multi-year visibility is not a cyclical memory stock. It's closer to a regulated utility — predictable cash flows, high barriers to entry, and contracted demand. Utilities trade at 15-20x earnings with 3-4% dividend yields. SanDisk is growing revenue at triple digits in its core segment. The market is applying a cyclical multiple to a secular growth story.
The AI storage scarcity premium: There are exactly zero other pure-play NAND companies with this scale and this level of AI exposure. Samsung is a conglomerate. SK Hynix is dominated by HBM. Micron is diversified across DRAM and NAND. If you want to express a view on AI inference storage demand, SanDisk is the only liquid, pure-play vehicle. That scarcity deserves a premium.
The HBF call option: The market is giving SanDisk essentially zero value for HBF. It's not in any revenue model. It's not in any analyst forecast. But if it works — and the technical approach is sound — it's a $10-20 billion revenue opportunity by 2028. That's a free call option embedded in the current stock price.
Now, the risks. I'm not going to pretend this is a risk-free trade. Here's what keeps me up at night:
Risk 1: The AI demand cliff (25-30% probability): If cloud providers hit a digestion period in 2026-2027 — if AI inference efficiency improves faster than expected, or if capex gets cut — NAND prices could reverse. The long-term contracts provide a floor, but they don't guarantee upside. My worst-case scenario is revenue down 20-30% and the stock trading back to 15x earnings.
Risk 2: The HBF execution risk (20-25% probability): This is a genuinely hard technical problem. TSV, advanced bonding, high-density interconnects — these are not trivial. If HBF samples fail to meet performance targets, or if customers choose HBM instead, the R&D spend ($500 million to $1 billion) is wasted. The technology is promising, but promising isn't shipping.
Risk 3: The competitive response (30-35% probability): Samsung and SK Hynix are not going to cede the AI storage market without a fight. They have deeper pockets, more advanced layer counts, and existing customer relationships. If they decide to flood the market with NAND capacity to maintain share, margins compress across the industry. The long-term contracts provide some protection, but not complete insulation.
Risk 4: The valuation reset (30-40% probability): At 30x earnings, the market is pricing in significant growth. If the 437% data center growth number decelerates to, say, 100% (which is still extraordinary), the stock could de-rate by 30-50%. The fundamentals are strong, but the entry price matters.
The Takeaway: What I'm Watching, and What I'm Doing
Here's my framework for the next 12 months:
The bull case: AI inference demand is not a one-year phenomenon. It's a structural shift in compute architecture. Every AI application — search, coding, agents, autonomous vehicles — requires storage. NAND is the substrate. SanDisk is the purest expression of that thesis. The long-term contracts provide revenue visibility. HBF provides optionality. The market is underpricing both.
The bear case: The memory industry has a 50-year history of boom-bust cycles. Every upcycle is followed by overcapacity and price collapse. SanDisk is a cyclical company wearing a growth stock's clothes. The 30x multiple is a trap.
My position: I'm long SNDK, but I'm sizing it as a 12-18 month trade, not a permanent holding. I'm watching three signals:
- The Q3 earnings report (November 2025): I need to see data center revenue growth hold above 200% year-over-year, and I need to see gross margin expansion. If those two things happen, the stock re-rates.
- NAND spot prices: I'm tracking TrendForce and DRAMeXchange data weekly. If spot prices start rolling over, the market will start pricing the next downcycle, and I'm out.
- HBF sample delivery: If SanDisk delivers samples to customers on schedule in 2026, and if there's any positive customer feedback, the stock gets a new valuation paradigm. If it slips, the optionality premium evaporates.
Here's the thing about Jane Street's 540% position increase: it's not a recommendation. It's a data point. But it's a data point that aligns with my own analysis of the order flow. The smart money is positioning for a repricing of AI storage. The question is whether you're positioned for it too.
I've been through the 2020 DeFi yield farming cycle, the 2022 Terra collapse, and the 2023 EigenLayer audit. I've learned that the biggest returns come from identifying structural shifts before the market fully prices them. SanDisk's transformation from a cyclical memory manufacturer to a contracted AI infrastructure provider is exactly that kind of shift.
The market is still debating whether AI storage is a real market. The $93.9 billion in contracts says it is. The 437% data center growth says it is. Jane Street's position increase says it is.
The only question left is whether you're going to act on it.
Based on my audit experience with storage protocols and my years of trading around semiconductor supply chains, I can tell you this: the setup here is real. The risk is real. But the asymmetry — the gap between what the market is pricing and what the contracts imply — is the kind of edge that defines a career.
I'm watching. I'm positioned. And I'm waiting for the market to catch up.
