• Counterpoint data shows enterprise SSDs reached 48% of global NAND bit shipments in Q2 2026, nearly double the 26% a year earlier, and this underscores the NBM as demand outstrips supply.

• But SanDisk's NAND revenue share has held flat at 12–13% for five straight quarters. Growth in this model comes from a rising market leading to rising ASP per bit, not from taking share.

• The onus is on SanDisk as to how it can control this demand-supply dynamic, which stays super tight into the back half of the decade on one end, and rising competition on the other from the likes of YMTC doubling down on investments and becoming cash-rich with an upcoming IPO.

• On the technology front, all eyes are on SanDisk’s Multi-Horizon innovations with BiCS9-13, HBF, 3D-Matrix Memory, eMMC leadership helping it stay afloat in all segments, Nanya equity stake to secure DRAM for eSSD portfolio, etc., which can help widen the gap to become a solid third-ranked player or more.

• Looking at the NAND Flash CAPEX / Incremental Output Intensity ratios puts SanDisk-Kioxia in a solid, disciplined position.

Memory: From Being Commodity to Being Not

SanDisk’s Investor Day, branded "In Focus," was a great pitch by SanDisk executives.

For three decades, NAND (like DRAM) has been valued as a commodity, and no amount of technology leadership changed that arithmetic. However, AI changed it. “Memory is the new Gold”

A 2Tb (256GB) die at ~80mm², thinned to ~35µm for 16-high stacking, is about 6.5mg of silicon. At the implied $0.50/GB (retail) from that drive price, the die is worth ~$128 — call it ~$20,000/gram, or roughly 140× gold. – Counterpoint Research

SanDisk framed its argument during the IR day very well, proving that the arithmetic has changed, it’s not a cycle anymore.

SanDisk is forecasting a FY2028 through FY2030 framework with:

• Revenue growing mid-to-high teens in line with bit growth,

• Non-GAAP Gross Margin sustained near 80%,

• Non-GAAP Operating Margin near 75% on operating expenses of roughly 5% of revenue,

• Adjusted free cash flow at about half of revenue.

CFO Luis Visoso framed it as optimizing for growth, sustainability and returns, with a commitment to return all excess cash to shareholders after reinvestment.

We think it was delivered straight and resonated well with the investor community. The stock closed up close to 14%, and the move pulled SK hynix, Micron, Western Digital and Seagate up with it.

New Wave Needs New Business Model

With this new AI wave from cloud to edge, the ASP for NAND Flash has gone up 4x in just 12 months, with a mix shift towards the server market where the demand has more than doubled. However, the bit supply has not gone up, causing severe scarcity and Chipflation.

As a result, SanDisk is leaning in to capture significant value in this AI wave via a newer structured model called NBM, or New Business Model. SanDisk believes it will help redefine its relationship with customers on different vectors — from allocation to pricing to financial guarantees — as NAND bits become more and more valuable.

Rather than selling into whatever the market pays each quarter, SanDisk now signs multi-year agreements (or LTAs) with committed volumes on both sides, enforceable minimum financial guarantees, and pricing that blends fixed and variable components with floors and ceilings. Supply and demand commitments are defined by year and by quarter. Eight customers have signed, covering approximately 50% of bits in FY2027 and roughly two-thirds in FY2028.

SanDisk describes this as becoming its predominant way of doing business.

If those contracts hold through a downturn, if any, which at this point looks difficult with AI tsunami, changes NAND into an asset. This can cushion the cyclicity if it turns out into one.

I believe this move to push cyclicity out of the equation is what can change the perception of investors, and it is a clever positioning and potential execution by SanDisk. If it can lock in two-thirds of FY2028 bits with an 80% gross margin target, that is great news for SanDisk and investors, and also for some of the marquee customers.

However, there could be a buried risk with increased concentration of customers and what it means for the long tail of customers and those relationships as the power has shifted. This could drive some difficult conversations and could limit growth unless the capacity increase is planned prudently.

The demand case checks out — for now

SanDisk is estimating 1.2 zettabytes of bits shipment consumption in demand for AI Datacenters alone by 2030, driven by AI inference and, specifically, by KV cache reshaping the memory hierarchy to Tier 3.5.

We talked about this in our NVIDIA GTC 2026 post with NVIDIA driving this right, front and center with NVIDIA BlueField-4 STX reference architecture to address CMX context memory storage for the Agentic AI wave.

Related: NVIDIA GTC 2026: Groq, Tokenomics, CMX

With Agentic AI, long context reasoning is elevating the role of NAND as inference storage intensity per token keeps climbing and KV cache offload to flash becomes de-facto architecture, in addition to Staging and enabling Faster Data Lakes.

SanDisk estimates the persistent KV Store Installed Base will jump to more than 1ZB by 2030.

According to Counterpoint’s NAND tracker and forecast, enterprise SSDs took 48% of global NAND bit shipments in Q2 2026, up from 26% a year earlier, as AI workloads migrated from training to inference. Server eSSDs are on track to absorb more than half of all NAND bits by the end of this year. That reallocation left consumer supply short and pushed consumer ASPs to record highs, with industry revenue rising fivefold year-over-year in Q2 2026.

SanDisk’s Datacenter revenue reached $2.98 billion in fiscal Q4 2026, up 103% sequentially from $213 million a year earlier, and up 437% for the full fiscal year. Fiscal 2026 revenue closed at $20.25 billion, up 175%.

Like the positioning of SSDs AS A TOKEN BATTERY

The Technology & Capital Efficiency is the real Differentiator

SanDisk has been one of the pioneers when it comes to solid-state storage technology leadership, with 19 generations - 11 planar (2D) generations followed by 8 commercial 3D generations developed. Though it has not completely translated into taking share away from the Korean peers, it is important to witness and acknowledge how the innovations have taken shape for and from SanDisk.

Mastering the Art of NAND Scaling

Building on CMOS directly Bonded to Array (CBA), SanDisk introduced what it calls two-dimensional scaling, which is the ability to mix and match array and CMOS generations to produce custom derivatives without a full node transition. SanDisk is trying to pack more bits per wafer by being creative with Architecture & Logical Scaling curves.

BiCS9 QLC is the first instance: a proven BiCS8 array paired with a BiCS10-based CMOS wafer, delivering AI-workload performance on already-depreciated array tooling.

Separately, BiCS10 QLC delivers a 60% bit density increase over BiCS8 with 65% more die per wafer. It’s a 332-layer 2Tb QLC CBA technology with a claim to be the world's densest memory chip.

With Kioxia, SanDisk has also shown a ninth-generation 2Tb QLC device running a 4.8 Gb/s NAND interface, a 33% step up.

SanDisk’s HBF – Great Invention but a Longer-Term Play

The hottest topic in the industry, though, is High Bandwidth Flash (HBF). And the billion-dollar question is: is it an HBM alternative? We believe not yet, as HBF is real optionality at this point, unproven for AI workloads and won't translate into revenue. It’s for sure a longer-term play. No commercial products or independent benchmarks exist yet, and nothing in the FY2028–FY2030 model appears to depend on it. If HBF lands, it is upside to the framework.

SanDisk and SK hynix published the first High Bandwidth Flash technical specification through the Open Compute Project on August 3 at FMS 2026, offering up to 512GB per device using 8-high and 16-high stacks, three bandwidth grades from roughly 0.4 TB/s to 3.0 TB/s, over a UCIe chiplet interface, with Google and Tenstorrent in the consortium. SanDisk described the ecosystem as gaining momentum.

3D Matrix Memory: From 3D Cross-Point to CXL, Bridging Gap between DRAM & NAND

SanDisk’s 3D Matrix Memory (developed in collaboration with research hub imec) is an emerging Storage-Class Memory (SCM) / next-generation persistent memory. It is designed to sit in the performance and cost gap between traditional DRAM and 3D NAND flash. From 2024 onward, SanDisk moved development to imec’s standard 300 mm pilot line and successfully delivered full 300 mm wafers and packaged test chips.

Traditional DRAM is facing severe physical scaling limits (sub-10nm nodes are very costly and complex). SanDisk is positioning 3D Matrix Memory to use a 3D cross-point/matrix array architecture to deliver up to 4× the density of DRAM at roughly 50% of the cost per bit. So some of the low-hanging use cases are AI data pipeline caching, context-window memory expansion, vector databases, and hyperscale cloud CXL memory pools.

Intel’s Optane proved the architectural demand for persistent memory but failed commercially due to proprietary ecosystem lock-in and high manufacturing costs. SanDisk is leveraging open standards (CXL) and standard 300mm wafer tooling to avoid those cost pitfalls. It does not compete directly with standard DDR5/HBM3e on raw peak bandwidth, but serves as an affordable memory tier for AI LLM inference pools and in-memory databases.

We believe 3D Matrix is still in the technology qualification phase. Moving from multi-gigabit demonstration to commercial mass-production chips (32–64 Gb+) will take at least 2-3 years. We will keep an eye on this development.

Optimizing Capital Intensity

What is also driving the 80%+ Gross Margins is SanDisk’s ability to decouple node migrations from interface performance, especially when the industry and SanDisk have been short of planned capital investments and cleanroom space due to the earlier cyclical nature of the industry. Though, I believe SanDisk-Kioxia will have to double down if they don’t want to leave money on the table.

SanDisk’s Market Position: Money on the Table!

SanDisk has been pinned between 12% and 13% of NAND revenue for five consecutive quarters according to the Counterpoint NAND Market Share Tracker. Over the same window, YMTC moved from 8% to 13%. SanDisk’s revenue and margin expansion has come from the market repricing, plus a mix shift into Datacenter, but not exactly from displacing competitors.

That is entirely consistent with the model SanDisk presented, which ties growth to bits rather than share. But it does mean the FY2030 outcome is largely a function of whether the market SanDisk participates in grows as management expects. On that, SanDisk’s own view is aggressive, estimating the NAND market above $300 billion in 2026 and above $500 billion in 2027, with output tight into 2028.

The competitive variable SanDisk said least about is the one we flagged in our Q2 analysis. YMTC entered the global top three by shipments at 14%, narrowly edging Kioxia, while mass-producing 267-layer 3D NAND and advancing 300+ layer technology on its Xtacking architecture. It ranked only fifth in revenue but darn close to the two above. Even though YMTC's mix remains consumer-weighted, and its China AI server business is constrained by US restrictions, it has stated plans to push that mix toward eSSDs in the second half of the year, with a listing widely expected between late 2026 and mid-2027. So YMTC, moving into the highest-value segment, is a supply-side risk that the NBM guarantees soften but do not remove.

Wrapping Up: Three Things to Watch

SanDisk made the strongest case at the Investor Day 2026, first on the business strategy - bringing in the NBM framework to prevent the business from any future cyclicity. Second, on the technology strategy and roadmap - CBA-based two-dimensional scaling, solid roadmap to lead with architecture and logic scaling, new initiatives from KV Cache alignment for AI DCs, HBF, 3D Matrix, to expand upon a capital-constrained decade. The demand shift toward inference storage is measurable and real and the next AI wave is going to be at the supply-constraint edge.

Three things we are watching whether the FY2028–FY2030 outlook holds:

1. Do NBM floors survive contact with a real downturn? Contracts with minimum guarantees have never been stress-tested in NAND. But still, we estimate it’s not going to happen, at least for the next eighteen months, so SanDisk can calibrate in the coming years if the time arises.

2. Does storage intensity per inference token keep rising and how much? The 1.2ZB TAM is just the tip of the iceberg. As token consumption shifts from maturing Generative AI to nascent-stage Agentic AI, it remains to be seen whether the market is severely underestimating demand. What happens once we're in 2028, when demand has moved even further ahead of supply? I believe whoever has invested in capacity by then will be the winner.

3. How fast does the right mix become crowded? As we noted last quarter, profitability through 2027 will be decided less by who ships the most bits than by who ships the right mix. SanDisk has the right mix today. So does everyone who can see the same data. But the wild card is the China-based peers – YMTC and, eventually, CXMT – getting into the NAND business.

For more NAND-market-related insights, from market share, shipment data, pricing and demand-side outlook, please reach out to Counterpoint Research's Memory & Storage Research. Please contact us for access.

Related reading

• Global NAND Memory Market Share: Quarterly

• Server-Led eSSDs Hit 48% of NAND Shipments; YMTC Enters Global Top Three