Business

Know the Business — Kioxia

Kioxia is the only listed pure-play NAND flash producer at global scale, and its P&L is essentially a leveraged bet on NAND average selling prices (ASPs). The company shipped more bits every year for the last five years, but operating margin moved from +22.7% to -23.5% to +37.2% — and the standalone March 2026 quarter printed 59.5%, with Q1 FY2027 guidance at ~74%. The case here sits between two unresolved questions: how much of the current earnings power is structural rather than a pricing event, and how durable the cost advantage embedded in the Yokkaichi/Kitakami joint venture with Sandisk really is.

1. How This Business Actually Works

The revenue equation is one line: (bits shipped) × (¥/bit). The cost equation is dominated by fab depreciation, silicon wafers, and electricity — almost all fixed once the fab is built. That mismatch — variable revenue versus fixed cost — is the entire reason the operating margin can swing 60 percentage points in 24 months on a business shipping more product every year.

Three implications professionals never forget:

• You are buying the pricing tape, not the bit count. Bits grew at high-teens % in every year — including FY2024 when Kioxia lost ¥253B at the operating line. The cycle is set by ¥/bit, not gigabytes.
• Operating leverage runs both ways and runs fast. FY2024 to FY2025: revenue +58.5%, operating profit swung from -¥253B to +¥452B — a ¥705B swing on a ¥630B revenue increase. That's because depreciation barely moved.
• The "moat" is shared. The Yokkaichi/Kitakami plants are jointly funded with Sandisk. Roughly 80% of plant capacity is 50/50 with Sandisk; only ~20% is exclusively Kioxia bits (per Integrated Report 2025). That structure halves the capex risk and halves the scale advantage relative to Samsung and SK hynix.

2. The Playing Field

Five firms control roughly 95% of global NAND bit supply. Kioxia is #3 standalone at ~14% share and roughly #1-equivalent when combined with Sandisk's 12% — they share the fab, so their bit cost curve is identical. Among listed peers, only Sandisk is a true economic substitute; the Koreans and Micron carry DRAM/HBM cushions and the spun-out WDC is HDD-only.

NAND share from TechInsights Q2 2025 / TrendForce / Morningstar Jun 2026. Peer multiples from Yahoo Finance / Fiscal.ai (as of June 2026). Kioxia operating margin and ROE are FY2025 IFRS reported. Fiscal periods differ across peers.

What the peer set actually reveals:

• Sandisk is the cleanest economic comparable and it is losing money at the same fab where Kioxia is printing 26.5% operating margins. The gap is product mix (Kioxia has more enterprise SSD), customer mix (Apple, Dell direct relationships vs Sandisk's retail/component skew), and timing of restructuring charges. The structural cost line is the same.
• Samsung trades at the lowest P/B (4.6x) and lowest P/E (26x) of the memory complex because its earnings are diluted by consumer electronics. SK hynix at 19.6x P/E is the HBM/AI-margin proxy — its multiple is rewarded for HBM dominance, not NAND.
• Micron's 26.1% operating margin in FY2025 — almost identical to Kioxia's — was driven mostly by HBM, not NAND. Read across: Micron's NAND-only business is still subsidised by DRAM/HBM. Kioxia's same margin is purer leverage to NAND ASPs.
• WDC is no longer a NAND peer. Post-Sandisk-spin (Feb 2025) it is a pure HDD vendor and now competes with NAND as an SSD-vs-HDD substitution lever rather than as a peer.

"Good" in this industry looks like: cost per wafer at the front of the cost curve, mix tilted to enterprise/data-center SSD, and a disciplined ability to underspend through the trough. Kioxia satisfies the first two through the JV scale and the data-center pivot. The third is unproven — Kioxia is a public company for only its second cycle.

3. Is This Business Cyclical?

Deeply cyclical, with the cycle hitting price first and margin second. Inventory and utilization swing alongside, but ASP is the single variable that explains 80%+ of margin variance over any 18-month window. Kioxia's own quarterly history is the cleanest demonstration possible: bit shipments grew almost monotonically; operating margin went from -23.5% to +59.5% (Q4 FY2026 standalone) in eight quarters.

Where the cycle actually hits, in order:

• ASPs — Q4 FY2026's revenue near-doubled QoQ on falling bit shipments, with the entire gain coming from price. This is the textbook NAND shortage signature.
• Working capital — trade receivables rose ¥422B in a single year. The balance sheet inflates faster than profit because customers wait 45-60 days to pay; in a downturn the same effect reverses and cash unwinds.
• Inventory — Kioxia's quarterly commentary in FY2025 and FY2026 explicitly references customer inventory normalization. Producer inventory days are not disclosed; this is one of the gaps the analyst fills with TrendForce data.
• Capex restraint — disclosed only with a lag through depreciation and through annual JV capex announcements. The Yokkaichi/Kitakami capex was up roughly 25% YoY for FY2026 — a signal Kioxia is leaning into the upcycle, exactly the behavior that historically reseeds the next glut.

4. The Metrics That Actually Matter

Generic semiconductor metrics — gross margin, P/E, R&D intensity — fail here. The metrics that explain value creation in NAND are mostly external, and the ones internal to Kioxia are not the income statement ratios.

Two metrics deserve a comment because they sit at the heart of the bull and bear cases:

• Bit growth vs market. Filings flag that Q4 FY2026 revenue gain came "partially offset by reduced bit shipment." If Kioxia is shipping fewer bits while ASPs explode, either it is allocating wafers to highest-margin customers (rational) or losing share to Samsung/SK at scale (concerning). The next two quarters will tell which.
• Customer concentration. Top three customers = 39% of FY2025 revenue, with Apple alone ~18%. The "Sandisk Group" 12% is a JV settlement, not an independent customer — it reflects the bit allocation mechanism. The genuine concentration is Apple + Dell ≈ 28%, and Apple has historically rotated suppliers between Samsung, SK, and Kioxia within a 24-month cycle.

5. What Is This Business Worth?

Value here is mostly determined by normalized through-cycle earnings power on the JV cost curve, not by current peak EPS and not by accounting book value. The right lens is mid-cycle EV/EBITDA layered with a separate read on JV co-ownership economics, not a single-year P/E. SOTP is not appropriate — there is one product line, sold globally, through one shared fab system.

A practical underwriting frame:

• At peak (current). Q1 FY2027 guidance implies an annualised non-GAAP operating profit run-rate near ¥5.2 trillion. Capitalize at any reasonable multiple and the implied value is enormous — but that is a peak-of-peak quarter, with explicit management language about ASP dependence.
• At a mid-cycle 15-20% operating margin on ¥2.0-2.2T of through-cycle revenue, you get ~¥330-440B non-GAAP operating profit. At 6-8x EV/EBITDA (cycle range for pure-play memory), the implied enterprise value is roughly ¥3-5T — small compared to the current share-price-implied EV but a useful anchor.
• The single most underwritten question is whether the JV cost curve and the mix shift have moved Kioxia from "memory cycle median" to "memory cycle leader." If yes, the right multiple is closer to SK hynix's 13x EV/EBITDA than to Samsung's commodity-blend lower-teens. If no, you reset to 6-8x and the current price implies a long peak.

6. What I'd Tell a Young Analyst

Three things. First, watch ¥/bit on TrendForce, not the income statement. The income statement is a 60-90 day lagged echo of the spot pricing. If the pricing tape rolls over, sell the stock the same week — do not wait for the earnings confirmation. Conversely, do not chase the stock after a record print; the record print is the confirmation that the easy money in the cycle has been made.

Second, the moat is real but it is shared. The Yokkaichi/Kitakami JV gives Kioxia a structural cost advantage that Samsung and SK can't fully replicate, and the 14% standalone share understates the company's effective scale. But ~80% of plant output is 50/50 with Sandisk, so any thesis that requires Kioxia to independently out-invest the Koreans is wrong. The interesting structural question is whether the long-rumored Kioxia/Sandisk merger or roll-up ever closes — that is the single transaction that would re-rate the cost curve into a market-share story.

Third, the biggest mistake is anchoring to FY2025 or FY2026 numbers as "normal." Kioxia has been public for 18 months. Its filings cover one trough (FY2024) and one peak (FY2026) and not a single regular year. Build your own normalized P&L, blend ~15-20% operating margin on ~¥2T revenue, run capex at 15-20% of that, and underwrite with explicit ASP scenarios. If the market is pricing today as the new normal, you are short. If the market is pricing the next downturn as 2024-redux, you are long. The cycle will eventually pass between those two priced extremes; that is where the alpha is.

What would change the thesis: a sustained ASP rollover with Samsung/SK adding wafer starts (bearish), a credible Kioxia/Sandisk JV consolidation that re-rates the cost curve (bullish), or a multi-year hyperscaler LTA at fixed pricing that converts the business from cyclical to contracted (bullish). Anything else is noise.