Industry

Industry — NAND Flash Memory and Solid-State Storage

SanDisk operates in one of the most cyclical, capital-intensive, and concentrated industries in technology: NAND flash memory. Five vertically integrated suppliers control essentially the entire global bit supply, fabs cost tens of billions to build, and pricing can swing 50%+ in a single year. The reader who understands the cycle understands the rest of this report.

1. What this industry actually does

NAND flash is non-volatile semiconductor memory — it stores data without power. It sits between two other storage technologies:

• DRAM (dynamic random-access memory): very fast, volatile, expensive per bit. Used as the working memory of CPUs and GPUs.
• HDD (hard disk drives): mechanical, slow, cheap per bit. Still dominant for low-cost bulk archive.
• NAND flash: persistent, fast (no moving parts), and steadily falling in cost per bit. Used in everything from microSD cards to enterprise SSDs.

The industry monetizes NAND in three main forms: wafers/components sold to other system builders, embedded modules sold into phones/laptops/cars, and finished drives (SSDs, SD cards, USB drives) sold to data centers, OEMs, and retail consumers.

2. The industry structure — five suppliers, oligopoly economics

Per SanDisk's own 10-K, the world's NAND supply is produced by essentially five vertically integrated manufacturers:

This is a classic oligopoly: high barriers to entry (fabs cost $10–20B), limited substitutes, and rational pricing behavior when discipline holds — but punishing price wars when it breaks. Each player's capex decisions ripple through ASPs for every other player 12–18 months later.

SanDisk's structural quirk is that it does not own any of its wafer fabs. Kioxia owns the seven (going on eight) Japanese fabs; SanDisk owns 49.9% of three operating Flash Ventures entities that lease the equipment and contract with Kioxia for wafer manufacturing at cost-plus-a-small-markup. Each partner is entitled to ~50% of output. SanDisk is fab-light on the front end but fully responsible for controllers, firmware, assembly/test, and go-to-market. Each side is contractually restricted from working with third parties to manufacture NAND while the venture operates.

Currencies are preserved as reported. The operating-margin column tells the cycle story far better than the scale column: at the same point in time, the DRAM-heavy peers (SK hynix, Micron) are clearing 25–49% operating margins driven by the AI/HBM boom, while pure-NAND SanDisk was still loss-making at the FY2025 close (June 2025) before the inflection arrived in calendar H2 2025.

3. The cycle defines everything

The single most important thing for a newcomer to internalize: NAND is a commodity cyclical, more like steel or oil than like enterprise software. Pricing, margins, and free cash flow swing wildly with the supply/demand balance — and the balance swings because supply is added in lumpy multi-billion-dollar fab steps with 12–18 month lead times, while demand from PCs, phones, autos, and data centers moves on its own clocks.

The last four years contain a full cycle in miniature: peak (FY2022) → trough (FY2023) → recovery (FY2024–FY2025) → AI super-cycle (FY2026).

The same shape shows up at every vertically integrated peer — the cycle is the industry-level variable, not a company-level one:

The mechanism is mechanical. When pricing is strong, every supplier expands capex. New capacity comes online 12–18 months later — typically into weaker demand. ASPs fall, and because fab depreciation and Kioxia JV fixed-cost obligations don't fall with them, gross margins go negative. Suppliers cut bit-output growth, take underutilization charges, and wait. Eventually, demand catches up to the smaller forward supply trajectory, ASPs rise, and the cycle starts again. The Q3 FY2026 print — gross margin of 78.4%, up from 22.5% a year earlier — is the up-leg of this same cycle, amplified by AI.

The shape on the right of this chart — gross margin tripling in two quarters — is the AI super-cycle ASP move expressing itself in real time.

4. End markets: data center is taking over the mix

SanDisk has historically reported revenue across three end markets, relabeled in FY2026 to track the new reality:

• Datacenter (previously "Cloud") — enterprise and hyperscaler SSDs. Smallest segment historically, now the fastest-growing and the structurally highest-margin home for the company's bits.
• Edge (previously "Client") — embedded NAND for phones, laptops, gaming consoles, automotive, IoT, industrial. The volume backbone.
• Consumer — retail SD cards, microSD, USB flash drives, portable SSDs. Defended by the SanDisk brand, slow-growing in unit terms and exposed to consumer cyclicality.

Datacenter went from $325M (FY2024) to $960M (FY2025), a 195% jump. Then on a quarterly basis it kept compounding:

Datacenter went from $197M in Q3 FY2025 to $1,467M in Q3 FY2026 — a 645% year-over-year jump. Management has signaled multi-year purchase commitments via what it calls "New Business Model" agreements (five signed by Q4 FY2026), explicitly designed to lock in the high-value AI-driven mix.

The geographic split underlines how globally distributed the demand is — and how concentrated the manufacturing is in Asia:

International sales were 80% of FY2025 revenue, almost all of it Asia-weighted because that is where the OEM and contract-manufacturer customers physically sit.

5. Cost structure, capex, and the Kioxia joint venture

NAND is a capital-intensive business in the structural sense, not in the sense that the income statement screams it. Capex-to-revenue at SanDisk has been only 3–4% in recent years — but that understates the real picture, because Flash Ventures sits off SanDisk's balance sheet and absorbs the bulk of the wafer-fab capital.

R&D consistently runs ~15% of revenue, in line with leading-edge semiconductor norms — the technology-node race (BiCS5 → BiCS6 → BiCS8 → BiCS9) consumes a lot of spend even before fab capacity is added. R&D stays roughly constant in dollars through the cycle, which is why it ballooned to 19% of revenue in the FY2023 trough.

What does the JV actually mean for the economics?

SanDisk pays Kioxia for wafers at cost plus a small markup. SanDisk also funds 49.9% of Flash Ventures' fixed costs — regardless of how many wafers it chooses to take. That last clause is the cycle whip in compact form: when SanDisk slows wafer purchases to align supply with demand, it still pays its fixed-cost share, and the unabsorbed manufacturing overhead lands directly in cost of goods sold. FY2024 saw $252M of underutilization charges; FY2025 saw $75M.

The Flash Ventures contracts run through 2029 and 2034. The structural relationship is not easily broken: each side is restricted from manufacturing NAND outside the JV. This is what makes the SanDisk–Kioxia tie-up de facto the third pole of the NAND industry, alongside Samsung and SK hynix.

6. Where we are now — the AI super-cycle

The current up-leg has a structural amplifier prior cycles did not: AI infrastructure demand. Three forces are working in parallel:

Force 1 — Enterprise SSD demand is exploding. Training and inference workloads need fast persistent storage attached to GPUs. Hyperscalers and AI infrastructure builders are buying enterprise SSDs in volumes that look nothing like the historical data-center procurement cadence. SanDisk's datacenter revenue went up 645% year over year in the latest quarter, with engagement reported across all five major hyperscalers.

Force 2 — HBM is absorbing memory peers' capacity. SK hynix and Micron are pouring DRAM wafer starts into high-bandwidth memory for AI accelerators. That tightens supply across the broader memory complex and supports NAND ASPs even before NAND-specific tightness sets in.

Force 3 — Supply discipline appears firmer than in past cycles. Capex announcements across the industry have stayed measured through the FY2025 trough. The big NAND makers all lived through the FY2023 wipeout. Whether discipline holds at full-cycle peak pricing is the open question every investor is asking.

The result inside SanDisk's own numbers:

The Q4 FY2026 guidance midpoint of $8.0B revenue at 80% gross margin would translate to roughly $6.4B of gross profit in a single quarter — more than SanDisk earned in all of FY2022 at the prior cyclical peak. The guided GAAP/Non-GAAP diluted EPS range is $30–$33.

7. Risks and regulatory backdrop

The industry's risk surface is shaped by where the fabs sit (Japan, Korea, Taiwan, China), where the customers sit (global), and how much the US government cares about who supplies whose chips.

Geographic concentration. SanDisk's wafer supply runs almost entirely through seven (going on eight) Kioxia-owned Japanese fabs. A natural disaster, contamination event, or extended power disruption affecting Yokkaichi could materially impair output. The 2022 contamination incident at Flash Ventures cost SanDisk $36M of recovered losses in 2024 — a small reminder of what fab-site concentration can deliver.

Tariffs and trade policy. Most of SanDisk's US-sold products are currently exempt from US tariffs, but management has flagged that loss of exemptions or new semiconductor-specific tariffs would compress US margins. International sales were 80% of FY2025 revenue.

China export controls. US restrictions on advanced semiconductor manufacturing equipment going to China are the main reason YMTC has not scaled to global-tier capacity. For SanDisk, this is a tailwind on supply but a headwind on China demand: the Unis Venture (48% SanDisk / 52% Unisplendour) is the China go-to-market channel.

Customer concentration. No single customer was 10%+ of revenue in FY2024 or FY2025, but one customer was 15% of revenue in FY2023. As the datacenter mix grows, customer-level concentration mechanically rises (there are only so many hyperscalers).

JV agreement risk. SanDisk is contractually tied to Kioxia until at least 2029 for the original ventures. Renegotiation, termination, or any restructuring of these agreements would have material consequences either way.

Global minimum tax (Pillar Two). International tax reform is expected to raise SanDisk's tax obligations starting in FY2026 as more jurisdictions implement the OECD-coordinated minimum-tax regime. The newly enacted "Big Beautiful Bill Act" (US, July 2025) reversed mandatory capitalization of US R&D expenditures but keeps the foreign capitalization rule in place — a net positive but with offsets.

8. Bringing it together — what the reader should hold in their head

First, the cycle is the dominant variable. Anything in SanDisk's revenue, margins, or cash flow that does not look mean-reverting on a 5-year horizon is either (a) the AI structural amplifier doing real work, or (b) the cycle's distortion that you should not extrapolate. Margins of 78% are not a normal-cycle steady state.

Second, SanDisk is structurally a pure-play NAND bet, not a diversified-memory bet. Micron and SK hynix get a margin shock-absorber from DRAM and HBM; SanDisk does not. That makes the upside steeper in the up-leg and the downside steeper in the down-leg. Comparisons against MU's ~26% FY2025 op margin or SK hynix's 49% are not apples-to-apples — those companies are taking part of their margin from HBM, which SanDisk does not make.

Third, the Kioxia JV is the most important non-obvious feature. It explains why SanDisk's on-balance-sheet capex looks low, why its fixed-cost absorption swings sharply with utilization, why it cannot meaningfully diversify wafer supply, and why a potential corporate combination between SanDisk and Kioxia would be the single largest restructuring catalyst the industry could deliver.