Thesis. 84.6% gross margin on a commodity, two-thirds of FY28 bits now contracted with minimum financial guarantees, zero debt and $14B of buyback — and NVIDIA confirming it cannot avoid paying the price. The strongest evidence set on the watchlist, against a supply wave with a known 2027 arrival date.
Reviewed 19 Sep 2026, after Q4 FY2026 results on 5 Aug and the 13 Aug Investor Day. Next review after Q1 FY2027 earnings on 5 Nov 2026.
What it does
Sandisk makes NAND flash memory — datacentre, client and consumer SSDs, plus embedded memory, cards and raw wafers. Spun out of Western Digital in February 2025. It does not own its fabs: manufacturing runs through Flash Ventures, a roughly 25-year joint venture with Kioxia operating fabs at Yokkaichi and Kitakami in Japan, extended through December 2034. CEO David Goeckeler. Beta 5.19.
The structural innovation is the NBM — "New Business Model" supply agreement. These are not ordinary purchase orders: committed volumes, enforceable contractual frameworks with minimum financial guarantees, and structured pricing with floors and ceilings, over 3–5 year terms with commitments defined by year and by quarter. Bernstein estimates the floor near $0.29/GB, broadly in line with Q2-2026 selling prices, and concludes they mute earnings downside into 2029–30 even in a collapse worse than 2010.
Most cyclical businesses have a floor you hope for. This one has a floor that is signed.
Bull case
- NBMs with eight customers covering ~50% of FY2027 bits and ~two-thirds of FY2028 bits (Investor Day, 13 Aug), with minimum financial guarantees. Management calls NBMs "quickly becoming its predominant way of doing business." The stock rose more than 10% on the day
- Q4 FY2026 revenue $8.965B, +372% YoY — beating the guidance top end by $715M and consensus by $403M. Non-GAAP gross margin 84.6% (up 6.2pts); non-GAAP EPS $39.25 against $30–33 guided
- FY2026 revenue $20.25B (+175%), GAAP net income $11.43B ($73.76 diluted EPS), non-GAAP diluted EPS $70.88. Datacentre +437%
- Sequential revenue growth was roughly one-third volume, two-thirds pricing — a quality improvement on the prior quarter, when bits fell high-teens sequentially and price did all the work
- Zero debt. $4.5B repurchased in Q4 (~89% of adjusted free cash flow) with $14B newly authorised, ~7.7% of market cap. Adjusted FCF excludes $1.938B of NBM prepayments — cash received that management chose not to count
- BiCS10 1Tb TLC in production since July 2026 at Kitakami Building 2: 332 layers, +59% bit density, +33% transfer speed, −34% power on data output versus BiCS8. That per-bit cost advantage is masked while prices rise and becomes the margin differentiator once the market stabilises
- High Bandwidth Flash (HBF) standardised with SK hynix, first OCP spec released 3 August — targeting 8–16x HBM capacity at comparable cost, for a tier where HBM is too expensive and SSDs too slow
- Long-term model through 2030: mid-to-high-teens revenue growth, high adjusted margins, all excess cash returned to shareholders
Bear case
- The supply response is funded and dated. SK hynix is investing ~$51B in a new NAND fab and the industry supply wave arrives from early 2027. Chinese capacity (CXMT, YMTC) is scaling with sub-commercial return requirements
- Q1 FY27 revenue guided $10.30–10.80B against $11.16B consensus — the ~5.5% shortfall that caused the selloff, even though EPS guidance of $44–46 bracketed consensus
- FY27 gross margin guided 83–85% — flat, not expanding
- It does not own its fabs. Manufacturing runs through the Kioxia JV, so unresolved reports of Chinese restrictions on NAND technology and tools hit production directly
- The NBM counterparties are not disclosed. This is the most important undisclosed fact on the page: price floors are only as good as the counterparty's credit and willingness to honour them
- Beta 5.19. It fell 55% in July alone — roughly $200B of market cap
- The market has de-rated this stock through three consecutive beats. A market that stops rewarding good news can keep not rewarding it, and a low P/E on peak memory earnings is a warning label rather than a discount
- Bears are right that double-ordering during shortages is how oversupply gets manufactured
Major customers
- Hyperscalers — datacentre SSDs, the segment that grew 437% in FY26. Individual names not disclosed
- Eight NBM counterparties — ⚠️ identities not disclosed
- Enterprise and OEM channel — server and storage vendors
- Client and consumer — PC OEMs, retail cards and drives
- Kioxia — JV partner in manufacturing, competitor in the end market
Read-through. Sandisk sits alongside MU in the memory layer, and its 84.6% gross margin is the clearest confirmation anywhere in this file that memory is currently the tightest link in the AI supply chain — compare AAOI at 27.7% or AMKR at 16.8%: same buildout, opposite pricing power. Three independent corroborations: NVDA guided its own gross margin down to 71–72% explicitly on memory prices; MU disclosed it can fulfil only 50–67% of demand; and AMKR blamed memory supply constraints for a Q3 communications decline — tight NAND and DRAM are now physically limiting production in adjacent industries. On the other side, P is paying these prices, and STX and WDC compete on the substitution curve as flash cost-per-bit falls.
What would change the view
- Gross margin direction. 84.6% is the high-water mark; any sequential decline is the cycle bell
- NBM coverage of FY27 and FY28 bits — the >50% FY27 trigger has been met, so the question is whether coverage extends further
- Buyback pace against the $14B authorisation
- Chinese restriction developments and Kioxia JV exposure — the manufacturing single point of failure
- HBF standardisation progress with SK hynix
- Bit shipment volumes — price without volume is not durable
- Whether the 2027 supply wave arrives on schedule, and at what utilisation
Smart money
- The company is the buyer. $4.5B repurchased in Q4, ~89% of adjusted FCF, with $14B more authorised
- Insider holdings rose ~8.5% quarter over quarter. No meaningful discretionary selling; the largest CEO "sale" was a Rule 16b-3(e) tax withholding of 1,569 shares. Goeckeler retains 509,903 shares
- Institutional base heavily index and quant weighted — a volatility amplifier in both directions
- Sell-side after Investor Day: BofA to $2,500 (from $2,100), Susquehanna to $3,250 (from $2,000) citing channel checks showing NAND pricing rising 75–100%. Targets span $1,750 to $3,250 across 14 analysts, median ~$2,000
Update log
19 Sep 2026 — Converted to the standard template and a stale claim removed. The page said "WDC remains a 10%+ holder, so further secondary sales are scheduled supply." That is no longer true — Western Digital completed the monetisation of its remaining 1.7M Sandisk shares in its fiscal Q4 and holds none. That overhang is gone. Also reconciled the NBM count: the Q4 call described ten agreements, the Investor Day described eight customers covering ~50% of FY27 and ~two-thirds of FY28 bits. The coverage figures are the current and more useful framing.
7 Sep 2026 — Verdict: Constructive → Conviction. Two-thirds of FY28 bits contracted with minimum financial guarantees converts this from a cyclical with a floor into something closer to a contracted business with cyclical upside.
13 Aug 2026 — Investor Day. NBMs with eight customers covering ~50% of FY27 bits and ~two-thirds of FY28 bits. Long-term model through 2030: mid-to-high-teens revenue growth and all excess cash returned. HBF framed as the answer to AI inference. Stock +10%.
5 Aug 2026 — Q4 FY2026 reported. Revenue $8.965B (+372%), beating the guidance top end by $715M; non-GAAP gross margin 84.6%; non-GAAP EPS $39.25 against $30–33 guided. FY2026 revenue $20.25B (+175%), datacentre +437%. $4.5B repurchased, $14B newly authorised. Q1 FY27 revenue guided $10.30–10.80B against $11.16B consensus; FY27 gross margin guided 83–85%.
3 Aug 2026 — First OCP specification for High Bandwidth Flash released with SK hynix.
Jul 2026 — BiCS10 1Tb TLC production began at Kitakami Building 2.
Research and education only — not investment advice.
Made with Bullet