Jane Street added 540% to its SanDisk (SNDK) position in Q2 2025. That's not a typo. Most retail eyes are glued to Nvidia's next move. The smart money is quietly stacking a different kind of AI infrastructure play.
Context: The Western Digital spin-off closed in February 2025. SanDisk emerged as a pure NAND flash IDM—design, fabrication, sales. No hard drives. No legacy drag. Just memory. The market initially treated it like a cyclical commodity play. PE swung between 20x and 25x. Then the $93.9 billion long-term supply agreement figure landed. Eight customers. Three major US cloud providers. Four to five years of locked-in revenue. The narrative shifted. But the price hasn't fully caught up.
Core: Let's audit the order flow. The typical storage cycle runs 2-3 years. Peaks and troughs. SanDisk's contract backlog breaks that cycle. It's a structural shift. I ran the math. Assume a 10% discount rate on the $93.9B backlog over 4 years. Present value is roughly $66B. Current enterprise value is about $45B. That leaves $21B in net present value for the un-contracted business. That's the HBF (High Bandwidth Flash) optionality, the enterprise SSD growth, and the smartphone recovery. The market is essentially getting the AI storage future for free if the core contracts are valued at 15x EV/EBITDA.
The Core Insight: The backlog transforms SanDisk from a spot-price dependent trader into a quasi-utility with upside. The 437% data center revenue growth guidance for fiscal 2026 is not a guess. It's pre-sold. The HBF product is the wild card. It's a new category for AI inference—not training. Inference requires high-bandwidth, low-latency access to large model parameters. NAND is faster than HDD, cheaper than DRAM. SanDisk is the first to market with a dedicated solution. Samples go to customers next year. If it gains traction, it's a new $10-20B market by 2028. If it fails, the R&D spend is a fraction of the contract value.
Contrarian: The consensus is that SanDisk is overvalued at 30x trailing earnings. They're right about the multiple. They're wrong about the denominator. The 2026 earnings estimate of $8.50 per share is based on the current product mix. It doesn't include HBF revenue. It doesn't assume any price increases beyond the locked-in contracts. The real risk is not the valuation. It's the cloud giants. They are designing their own storage controllers. AWS has Nitro SSD. Google has its own flash translation layer. If they move to full vertical integration, SanDisk's long-term contract renewal could be at lower margins. But that's a 2028 problem. The contracts are signed. The cash flows are coming.
Contrarian Angle: The HBF hype is a distraction. The real value is the boring backlog. The market is pricing in a 40% revenue growth for the next two years. The backlog guarantees 30% of that. The rest is dependent on spot market strength. If AI demand slows, the spot price drops, and the high-multiple collapses. But the backlog provides a floor. The stock is a downside-protected call option on AI inference.
Takeaway: The low is the $80 level. The high is $150 if HBF hits. The smart money is accumulating. The market doesn't care about your thesis. It cares about cash flows. The cash flows are locked. I bought the silence between the candlesticks.