Layer 3 · SNDK — SanDisk Corporation
One-line thesis
SanDisk is the cheapest NAND pure-play that just printed 78% gross margin on +250% YoY revenue — own it as the AI-storage-tightness call and accept that the parabola will eventually correct.
What SanDisk physically does
SanDisk fabricates 3D NAND flash memory — the storage substrate that sits one tier below DRAM in the memory hierarchy and absorbs the cost-sensitive bulk of AI training datasets, model checkpoints, agentic-AI memory, vector databases, and the enterprise SSDs that increasingly displace HDDs at the warm-tier of hyperscaler storage. Post-separation from Western Digital in early 2025, SanDisk is now the standalone NAND pure-play, jointly operating the Yokkaichi and Kitakami fabs with Kioxia under their decades-long manufacturing JV. Physically, SanDisk's wafers are produced at 100+ layer 3D NAND stacks, packaged into enterprise SSDs, client SSDs, and the new HBF (High Bandwidth Flash) form factor that the Kioxia-SanDisk roadmap is positioning as the AI-inference storage tier.
The AI stack relevance is that storage tightness has finally arrived at NAND. For most of the past four years AI capex was bottlenecked at GPUs, then at HBM, then at advanced-packaging substrates. NAND was the spare-capacity layer — over-supplied, low margin, the loss-leader of the memory complex. That has flipped. AI inference workloads at scale require enormous volumes of warm-tier storage, agentic AI workflows require persistent memory for context and tool-state, and the migration of vector databases onto SSD-class storage has pulled enterprise NAND demand to levels the industry never planned capex for. SanDisk and Kioxia, both running the same Yokkaichi/Kitakami fabs, are the principal beneficiaries.
The HBF roadmap is the AI optionality. HBF is a NAND-based, HBM-form-factor flash module designed to sit on the same package as the GPU and serve as a 1-2 tier slower but 10-100x cheaper-per-bit substitute for HBM-only architectures in inference workloads. If HBF clears the qualification cycle at Nvidia or AMD for any production inference SKU, it is a step-change re-rating event for the joint SanDisk-Kioxia franchise.
The mechanics of why NAND is suddenly tight in 2026 are worth spelling out because most generalist investors still think of NAND as the perennial-oversupply layer of the memory complex. Three things changed simultaneously. First, the industry under-invested in capex from 2022 through early 2025 — Samsung, Hynix, Kioxia, Micron and SanDisk all cut node-transition capex during the 2022-23 downcycle, which means 2026 supply is structurally below trend. Second, AI inference workloads create a demand pull on enterprise SSD volumes that nobody modelled — every agentic AI deployment, every RAG system, every long-context inference cluster needs persistent storage at multi-petabyte scale. Third, the substitution of HDD warm-tier with SSD at the hyperscaler tier is no longer a 2027-28 story but a 2026 story, because the GB-per-watt economics finally crossed over in the high-density nearline use case. All three pull bias up on the demand curve at the same time supply curve under-shoots. That is the NAND supercycle.
The financial print
Q3 FY26 (March quarter, reported April 30) was a blowout — revenue $5.95 billion, +250% YoY, EPS $23.41 versus $14.62 consensus (+60% surprise), non-GAAP gross margin 78.4% — a 5,570-basis-point year-over-year expansion. These are not normal memory numbers. The 78% gross margin print is what an AI accelerator looks like at peak pricing, not what a NAND fab has historically delivered.
The mechanism is pricing leverage. SanDisk disclosed NAND pricing +130% QoQ and +200% YoY — the steepest unit-price increase in NAND's modern history. Capacity is sold out, customer LTAs are being renegotiated higher into 2027, and the cost base is structurally lower than the FY24 trough because the fab JV has been running depreciation-heavy. The stock has closed at $1,381 on May 17 — up 482% year-to-date — and prints all-time highs as the analyst community walks consensus higher. Q4 FY26 reports July 30.
Customer mix
Customer concentration is hyperscaler-skewed but more diversified than the HBM names. The top five customers — AWS, Microsoft, Google, Meta and one tier-1 OEM — likely account for 50-60% of revenue, with the enterprise SSD line carrying the bulk. Nvidia is an indirect customer through the AI-server OEM channel; the direct relationship is still developing on the HBF roadmap.
The product-mix evolution inside the customer book is what supports the through-cycle margin step-up. Through 2023-24, SanDisk's revenue mix was roughly 60% client/consumer NAND (mobile, PC, USB sticks, SD cards) and 40% enterprise NAND. Through 2026, that mix has flipped to roughly 35% client / 65% enterprise as the hyperscaler SSD demand has compounded. Enterprise NAND carries 1,500-2,500 basis points higher gross margin than client NAND structurally, so the mix-shift alone explains a meaningful slice of the 78% gross margin print without invoking commodity-pricing peaks. The mix is durable — enterprise SSD demand does not mean-revert the way client NAND does — and that supports the bull-case argument that through-cycle margins are now structurally higher than the historic peer comp.
The customer LTAs are the key signal. SanDisk has explicitly disclosed that NAND pricing is locked into 2027 at escalators above current spot, that capacity for the next 12-18 months is contracted, and that LTAs into the back half of the decade are being negotiated. This is a meaningful change from the historic NAND contract structure, which was effectively spot pricing on a quarterly basis. The customer mix is also tightening upward — hyperscalers are now willing to pay premium for guaranteed allocation rather than chase spot, which is itself the indicator that the cycle is structurally different.
Competitive context
The NAND market has consolidated to a five-vendor structure: Samsung, SK Hynix/Solidigm, Kioxia, SanDisk and Micron — and effectively two technology consortiums (Samsung-Hynix and Kioxia-SanDisk via the JV). SanDisk is the western-listed wrapper for the Kioxia-SanDisk fab capacity, which represents roughly 30-35% of global NAND supply. The moat is the JV manufacturing scale, the existing customer qualifications, and the technology roadmap (BiCS 9 and BiCS X high-layer 3D NAND nodes).
The capex-discipline dynamic across the NAND vendor group is what makes the current pricing power sustainable rather than transient. Through 2022-23, Samsung dominated capex expansion and effectively over-supplied the market, which is why the 2022-23 NAND downcycle was so severe. Through 2024-26, Samsung has been more disciplined, Hynix has been focused on HBM rather than NAND, and the Kioxia-SanDisk JV has prioritised cash returns over capacity expansion. The result is industry capex running below the demand growth curve, which is the structural condition for sustained pricing power. Watching the Samsung and Hynix NAND-specific capex commentary at each print cycle is the key forward indicator.
SanDisk's competitive position has improved markedly since the separation from Western Digital. As a standalone NAND pure-play with a tighter cost structure and direct hyperscaler relationships, the operating leverage profile is now clean — no HDD-cycle drag, no portfolio-discount overhang. Against Micron's NAND segment and Samsung's NAND business, SanDisk is the marginal-cost beneficiary because the Yokkaichi/Kitakami JV depreciation is largely behind it, while Samsung is still running large new-fab capex. The competitive context is favourable through the cycle peak and probably for another 18 months after.
The market-share data is informative. SanDisk's standalone NAND share is roughly 14-16% globally; combined with Kioxia inside the JV the manufacturing footprint controls 30-35% of global capacity. Samsung remains the share leader at ~32%, SK Hynix/Solidigm runs ~17-19% combined, Micron ~10-12%. The competitive battleground in the up-cycle is principally the enterprise SSD segment, where the SanDisk-Kioxia JV is gaining share on the back of BiCS-class technology that outperforms Samsung's V-NAND on per-watt at the highest capacities. The enterprise SSD share gain is what is fuelling SanDisk's revenue out-performance versus the sector — the company is taking share inside a tight market, which compounds the pricing-power lever.
Terminal risk
The terminal risk is the parabolic move itself. +482% in five months is a sign that the marginal buyer is paying any price for the narrative, and when AI capex visibility blinks — whether for one quarter or one cycle — the unwind is brutal. Historic NAND cycle peaks have given back 50-70% of the upcycle move within 12 months. The risk to the thesis is not that the cycle is wrong; it is that 482% has already priced the cycle. A capex blink at any major hyperscaler in the 2027 print cycle is the trigger.
A secondary technical risk is the Chinese NAND capacity overhang. YMTC's domestic Chinese fab capacity is ramping; while the absolute volume is still small in global context, the trajectory through 2027-28 could meaningfully add bits at the low end of the market, particularly for consumer and client NAND. The displacement risk is at the volume tier rather than the enterprise tier where SanDisk is concentrating, but any pricing pressure on commodity NAND ripples through the wider market and could compress SanDisk's blended ASP.
Bull case
The base bull case is structurally tight NAND through 2027, HBF qualification at one hyperscaler-class customer by end-2026, and FY27 revenue running $22-25 billion versus $18-19 billion FY26 consensus. EPS in the $90-110 range at peak, a 12-15x multiple — a NAND business with HBF optionality and a tight supply curve probably deserves a higher multiple than the historic peer math suggests — and you arrive at $1,400-1,700 as a base bull range.
The stretch bull case adds HBF being a real GPU-package memory tier by 2028, which materially re-rates the entire Kioxia-SanDisk franchise on a sum-of-parts and pulls the multiple to high-teens. The implied upside in that path is $2,500+ over three years, but it is a probabilistic call rather than a base case.
The capital-return scenario is the underappreciated leg of the bull case. SanDisk's FY26 free cash flow is running at $4-5 billion against a market cap of roughly $85-90 billion at $1,381 — a free-cash-flow yield in the high single digits at peak earnings. The board has signalled openness to buybacks once leverage targets are met (the post-separation balance sheet was deliberately structured for capital flexibility), and a sustained buyback at $1,200-1,500 would compound earnings-per-share growth through the cycle peak in a way that pure-volume models do not capture.
Gap / bear case
What the market may be missing is the duration risk on the upside — if NAND ASPs hold or push higher into 2027 because hyperscalers continue to pay for allocation rather than wait for capex to land, the FY27 number is dramatically above consensus and the stock can grind higher despite the 482% YTD move. Conversely, the bear case the market may be under-pricing is operating-margin mean reversion: 78% non-GAAP gross margin is a peak print, not a through-cycle number, and a 1,000-bp compression in FY27 takes the EPS down 25-30% even on flat revenue.
The market is also probably under-pricing the JV-structure risk. Kioxia and SanDisk co-own the fabs; any disagreement on capex pacing, technology roadmap or HBF prioritisation between the two parents creates execution friction. The post-IPO Kioxia has more strategic flexibility now, which is a positive for HBF velocity but a complication for joint-capacity planning.
The capex-cycle timing is also worth highlighting. Through 2026, the JV is running fabs near full utilisation with limited capex expansion; from 2027 onward, the JV may need to commit to a new capacity phase to support continued demand growth. The capex commitment will compress free cash flow temporarily but is required for the bull-case revenue path. Watching the JV capex commentary at the Q4 FY26 print and 2027 capex guidance will be the leading indicator of whether SanDisk is prioritising capacity expansion (revenue-positive long-term, FCF-negative short-term) or maximising cash return (FCF-positive short-term, revenue-cap long-term).
Optionality
Three options. First, HBF — the GPU-package flash form factor — is the single biggest free option in the storage complex. Second, the standalone listing structure gives SanDisk strategic flexibility to spin out the JV stake, do a sale-leaseback on fab capacity, or initiate a buyback that the legacy WDC-attached structure could not. Third, enterprise SSD displacing HDD warm-tier storage — the secular substitution — is accelerating, and at current NAND $/GB the crossover economics are reaching the point where the next leg of HDD-to-SSD migration is on the table for 2027-28.
The trade
Entry: $1,280-1,400 — acceptable on the current melt; add on any 15-20% pullback to $1,150. Size: 3-4% portfolio target; treat as a torque sleeve rather than a core. Stop: $950 (200-day MA / prior breakout retest). Catalyst date: Q4 FY26 print July 30, 2026. Trim/exit: trim 25% at $1,800, 50% at $2,100; full exit on NAND ASP roll-over in monthly Trendforce data. Conviction: 8/10.