Industry

Industry — NAND Flash Memory

Figures converted from JPY at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

NAND flash is a global commodity oligopoly: five producers (Samsung, SK hynix/Solidigm, Kioxia, Sandisk/Western Digital, Micron) sell almost every bit shipped. The same gigabyte sold for ~80% less three years ago and ~40% more three months ago. Profits here are not a function of bits shipped — they are a function of how the bit price moved during the quarter. The frame to avoid: Kioxia is not simply a "growth" story tied to AI. The recent earnings explosion is a pricing event layered on structural bit growth, and the same pricing mechanism that produced $3.74B of Q4 FY2026 operating profit produced a $1.68B operating loss twenty-one months earlier.

1. Industry in One Page

NAND flash memory is non-volatile storage — it holds data without power — and sits inside every smartphone, SSD, USB drive, automotive infotainment unit, and data-center server you can name. The world buys flash by the gigabyte. Sellers compete on cost per bit (driven by 3D layer count, lithography, and wafer scale) and on product mix (enterprise SSDs > client SSDs > mobile NAND > raw die). The market is concentrated, capital-intensive, and famously cyclical.

Sources: Mordor Intelligence NAND Flash Memory Market Report (2026-2031); TechInsights NAND Market Report Q2 2025 (via Kioxia Integrated Report 2025); TrendForce Q2 2024 share tracking; Morningstar 285A profile (Jun 2026).

The trap to avoid. A NAND analyst who watches only bit shipments will mis-read every cycle. Bit demand has grown at a high-teens CAGR for a decade with little volatility, but industry revenue swung from US$53B (2021) to US$39B (2023) to a forecast ~US$72B (2025). The signal is $/bit (ASP), not GB shipped. The 2023 trough margins and the 2026 peak margins both happened against fairly steady underlying bit growth.

The chart below is the same point in Kioxia's own P&L: the company shipped more bits every year, but operating margin moved between -24% and +37% — and the standalone quarter ended March 2026 printed 59.5%.

2. How This Industry Makes Money

NAND producers sell bits, not chips. The economic unit is $ per gigabyte (or per bit), and the revenue equation reduces to (bits shipped) × (ASP per bit). Cost is dominated by depreciation of multi-billion-dollar fabs, silicon wafer cost, electricity, gases, and R&D. Variable cash cost per wafer is comparatively small once the fab is built; that is why the industry's gross margin can swing 40 percentage points in 18 months without touching headcount.

Three structural features control the profit pool:

• High fixed-cost intensity. Capex runs at 15-25% of revenue across the cycle and depreciation is the largest single cost item. When ASPs fall, producers cannot cut cost fast enough, so the operating-margin line whipsaws.
• Scarce true substitutes at the storage tier. Hard-disk drives still win on raw $/TB for cold storage, but SSDs displace HDDs in nearly every active-data workload; SSD-vs-HDD substitution is now the dominant secular tailwind, with HDD supply visibly tightening (TrendForce flags ongoing HDD shortages pushing NAND prices in 4Q25).
• Power sits with whoever has scarce capacity. In upcycles, producers can price at will because hyperscalers cannot wait 18 months for a new fab. In downcycles, the largest customers (Apple, Dell, Sandisk, Samsung's own assembly) extract concessions because producers cannot let utilization slip.

Two implications. First, the equipment makers earn the most stable profits across the chain — they sell to every NAND producer regardless of who is winning that quarter. Second, NAND fabs themselves are the cycle's risk-bearer: they absorb both the upside and the downside of the gap between bit supply and bit demand, which is why memory pure-plays trade on EV/EBITDA at peaks and book value at troughs.

3. Demand, Supply, and the Cycle

Demand is structurally growing at roughly 20% per year on a bit basis. Supply ought to track demand — and over five-year horizons it does — but each new wafer line takes 18-24 months to bring online, so over any 12-month window supply is approximately fixed and demand shifts dominate. The cycle is then governed by inventory, hyperscaler order patterns, and producer discipline on capex.

Supply is governed by layer count (more 3D layers per wafer = more bits per wafer = lower $/bit), wafer additions (slow, expensive, and politically visible), and node migrations (each transition temporarily depresses yield and bit output). Kioxia is migrating from 8th-generation BiCS (218L) toward a 10th-generation product offering ~59% higher storage density on a similar wafer footprint, and the Kioxia/Sandisk JV's reported capex is up ~40% YoY for FY2026 (TrendForce, June 2026).

The cycle shows up first in ASP (visible in TrendForce / DRAMeXchange spot pricing within days), then in inventory days at producers (visible in the quarterly balance sheet two months later), then in utilization (only disclosed by management commentary), and last in operating margin (the quarterly P&L). A reader watching ASPs in real time has roughly a one-quarter lead on the margin print.

4. Competitive Structure

NAND is a five-firm global oligopoly with a sixth Chinese entrant (YMTC) handicapped by US export controls. The top three producers control roughly two-thirds of bit supply. Kioxia and Sandisk are economically siamese: they share Yokkaichi (Mie, Japan) and Kitakami (Iwate, Japan) production through the "Flash Ventures" / "Flash Forward" joint ventures, so their reported shares are partly double-counted at the fab level even though their commercial channels are independent.

Share estimates blend TrendForce Q2 2024 quarterly tracking (Samsung 36.9%, SK Group 22.1%, Kioxia 13.8%, Micron 11.8%, WD/Sandisk 10.5%) with Morningstar (Jun 2026) and Bitget industry summary 2024-25 ranges. Combined Kioxia + Sandisk approaches 26-27% of bit supply via the Yokkaichi/Kitakami joint output.

Three structural features matter for an investor.

• Concentration is genuine, but discipline is not. Five firms control supply, yet the industry has historically been unable to coordinate capex through downcycles, which is why margins still go negative.
• Korea has scale; Japan has technology. Samsung and SK Hynix run more wafer starts, but Kioxia's per-wafer cost is reportedly lower (UBS, May 2026) because of lateral cell scaling rather than pure vertical stacking. Whether that cost advantage holds at 400+ layers is the structural question for 2027-2028.
• The JV structure is unusual. Kioxia and Sandisk operate the same fabs but go to market separately. A formal Kioxia/Sandisk merger has been floated repeatedly; SK Hynix is the most vocal antitrust opponent (Mordor Intelligence, 2026).

5. Regulation, Technology, and Rules of the Game

NAND is one of the most politicised industries in semis. The rules an investor must hold in their head:

The single most important regulatory fact for a NAND investor is the US cap on Chinese leading-edge NAND tools. Without it, YMTC would already be a top-three producer and the supply curve would be materially steeper. The single most important technology fact is CBA / hybrid bonding plus QLC, because together they extend the cost-down curve another two product generations and prolong the SSD-vs-HDD displacement runway.

6. The Metrics Professionals Watch

NAND analysts ignore most generic ratios and watch a small, very specific set of indicators. Most are not in the income statement; they live in industry trackers (TrendForce, DRAMeXchange, TechInsights) and in management commentary.

Pro tip for the reader. ASP and inventory days lead the operating margin print by roughly one quarter, sometimes two. If you only watch the income statement, you are always looking backwards.

7. Where Kioxia Holdings Corporation Fits

Kioxia is the #3 global NAND producer (~14% standalone, ~26-27% combined with Sandisk through the joint fabs) and the only pure-play NAND scale producer of the top five. Samsung and SK Hynix are diversified memory + semi conglomerates; Micron is DRAM-led; Sandisk is a US-listed pure-play that shares Kioxia's fabs. That means Kioxia's earnings have the highest beta to the NAND cycle of any listed peer — there is no DRAM or HBM cushion when bits get cheap, and no other product line to dilute the upside when they get expensive.

Kioxia is the highest-beta liquid expression of the NAND cycle on global exchanges, with a credible (and government-co-funded) capex roadmap and a real cost-curve story, but with no internal hedge against the next downturn other than its JV structure and balance-sheet management.

8. What to Watch First

The industry backdrop for Kioxia improves or deteriorates faster than the income statement. Five to seven signals worth watching, in roughly the order they move:

If pricing weakens before Samsung/SK announce capex restraint, the cycle is rolling over. If hyperscaler long-term agreements multiply and YMTC remains tooling-constrained, the upcycle has more room. The reported numbers print 60-90 days behind the TrendForce tape.