Deck
Kioxia is the only listed pure-play NAND flash memory maker at global scale, spun from Toshiba in 2018 and IPO'd on Tokyo Prime in December 2024. NAND wafers come from a 50/50 joint venture with Sandisk at the Yokkaichi and Kitakami fabs.
Trailing P/E 76× says expensive. Forward P/E 8.2× says cheap. Both are right.
- The print that decides it. Management has guided Q1 FY27 (ending June 2026, reports August 7) to ¥1.75T revenue and ¥1.30T operating profit — a 74% margin, the largest single-quarter guide a NAND maker has ever issued. At guide, FY27 EPS lands near ¥9,500 and the forward multiple drops to 8.2×, below SK hynix's 19.6×.
- What can't annualize. Management's own published long-term financial model targets a mid-20s operating margin through cycle, against Q1's guided 74% and FY24's printed -23.5%. Normalize at 15-20% on ¥2.0-2.2T revenue and FY27 EPS lands closer to ¥4,500 — half what the forward multiple is built on.
- Why the gap matters. Q4 FY26 alone delivered 43% of full-year revenue (¥1.0T of ¥2.34T) and 68% of full-year operating profit. The whole bull case is whether that ¥1.0T quarter annualizes; the whole bear case is that it was a single ASP event being capitalized into perpetuity.
NAND ASPs swing the P&L by 60 points of operating margin while bit shipments grow every year.
NAND is a fixed-cost commodity sold by the bit. Revenue is bits × ¥/bit; the cost stack is wafer depreciation, electricity, and silicon — all locked in 18-24 months before the bit ships. In two years bit shipments grew at the industry rate, ASPs near-doubled, and operating profit swung ¥1.1T on a ¥1.3T revenue increase. The same mechanism runs in reverse: in FY24, with bits still growing, ASPs collapsed and Kioxia lost ¥253B. The next downturn will look the same — the only question is when.
Receivables tripled in a single year. Whether the super-quarter is billings or cash decides the whole forward multiple.
- Receivables ¥238B → ¥660B in twelve months. DSO doubled from 51 to 103 days; the FY26 working-capital drag on operating cash flow was ¥398B. Receivables grew 177% YoY while revenue grew 37% — a gap too wide for ordinary quarter-end timing.
- Q4 carried the entire build. Q4 FY26 alone was 43% of full-year revenue, billed in the final weeks of the fiscal year. Whether the cash collects in H1 FY27 is the single binary the forensic case names as decisive. If receivables hold above ¥660B at the half, the 8.2× forward multiple is built on an EPS denominator that overstates real economic earnings.
- ¥395B goodwill from the 2018 Toshiba carve-out sits on the balance sheet untested through a NAND down-cycle since IPO — 11% of total assets that stands or falls on the same Q4 cash quality.
Bain cut 51.6% → 27.7% in six months while paying ¥0 common dividend across three fiscal years of record profits.
- The exit pattern. Bain Capital monetized at ~50× IPO price through a chain of Goldman-led blocks — from 51.64% in November 2025 to 27.69% by May 2026. On May 26, Bain filed it "may make material proposal on director selection" ahead of the June 25 AGM. The most NAND-cycle-experienced seller in the cap table is leaving the table.
- ¥0 common dividend on ¥554B FY26 net income. Three fiscal years of zero cash to ordinary shareholders, including the FY26 peak. The FY27 "progressive dividend up to 50% of net cash flow" is a promise — first declaration depends on the AGM, where Bain's residual 28% can still dominate any contested vote.
- CEO Hayasaka and CFO Hanazawa both turning over in the same quarter as the US ADS listing filing. Textbook peak-cycle PE exit script: replace executives, widen the eligible-buyer pool, trim into the new bid.
Two companies, same fab, same per-wafer cost. Kioxia FY25 op margin +26.5%. Sandisk FY25 op margin -18.7%.
The cleanest moat test in the dossier. Identical wafers come off the Yokkaichi/Kitakami JV split 50/50. In FY25, Kioxia printed a +26.5% operating margin; JV partner Sandisk printed -18.7% on the same silicon. A 45-percentage-point spread on shared cost.
What's structural, what's not. UBS estimates Kioxia per-wafer cost at roughly half of Samsung and SK hynix, and METI is co-funding ¥243B of the ¥729B next-stage capex — a ~33% effective capex haircut Korean peers do not get. But the cost is shared 50/50 with Sandisk; the 45pp spread is enterprise SSD mix and direct Apple/Dell/hyperscaler channels — both of which Sandisk has named as strategic priorities.
What it didn't protect. On the same JV cost curve celebrated today, FY24 printed -23.5% operating margin. The moat is a cushion against peers in a down-cycle, not a shield against the cycle itself. Kioxia has never been a public-market dividend payer through a NAND trough.
Lean watchlist — the cyclical setup is real, but the informed-seller tape and the receivables build are too loud to chase ¥78,140 ahead of confirmation.
- For. Goldman models NAND deficits of -4.4% / -4.6% / -3.0% through 2028; FY26 capacity is already sold out; Apple accepted a ~70% NAND unit-price hike under quarterly renegotiation. The structural cycle is real and one print from confirming the forward multiple.
- For. The Yokkaichi JV cliff was retired — Sandisk paid $1.165B to extend the agreement through 2034 — and S&P/Fitch upgraded Kioxia to BBB- in May 2026 with preferred shares retired. Every pre-IPO risk pillar named by FY24 bears has been retired in eighteen months.
- Against. 74% Q1 op margin cannot annualize against management's own "mid-20s through-cycle" long-term model and an FY24 trough of -23.5%. Normalized FY27 EPS lands closer to ¥4,500 than ¥9,500, and the 8.2× forward multiple re-rates.
- Against. Receivables tripled to ¥660B with DSO at 103 days, Bain sold 51.6% → 27.7% on ¥0 common dividends across three fiscal years, and ¥395B of carve-out goodwill has never met a NAND down-cycle. These are durable thesis variables — they outlast the August 7 print.
Watchlist to re-rate: Q1 FY27 print on August 7 — op margin and the receivables QoQ delta. June 25 AGM — first common dividend declaration and the disposition of Bain's "material proposal." TrendForce monthly contract NAND ASP — two consecutive QoQ declines would invalidate the deficit-through-2028 premise.