BiCS, Jointly OwnedNarrow moat

Sandisk (SNDK) — moat facet

A supplier can cut you off; a co-owner cannot. That single distinction is why a company with no fab and 12.4% of the market is not at anybody's mercy.

The chips are called BiCS, and neither Sandisk nor Kioxia can be cut off from them, because they own them together.

Second on the only scoreboard that matters in a fungible product — and the technology is co-owned, so it cannot be taken away.

The arrangement is stated plainly in Sandisk's filing: the two companies co-develop flash technologies, including process technology and memory design, for Flash Ventures' use, and jointly own the co-developed technologies. Each also licenses to the other the technologies it developed independently that are reasonably necessary to the joint work.¹

That is a materially different thing from a supply agreement. A supplier can raise your price, deprioritise your orders or refuse to renew. A co-owner of the design cannot take it away from you, because the design is as much yours as theirs. It is the reason Sandisk — a company with no fab and 12.4% of the market — is not a merchant buyer at the mercy of whoever has spare wafers.

The technology itself is competitive without being ahead. BiCS8 is the eighth generation at 218 layers, using CMOS directly bonded to array, with roughly 50% higher bit density than its predecessor. BiCS9 keeps those cells and adds a faster separate logic layer. BiCS10 is 332 layers, built by stacking three 100-plus-layer strings rather than as a monolith, with a 4.8 Gbit/s interface about 33% faster than BiCS8; both partners are sample shipping and mass production is expected in 2027 at Kitakami.² For comparison, SK hynix is at 321 layers, Micron at 276, and Samsung at 400.³

Rated narrow. The joint ownership is genuinely durable and genuinely unusual; the technology position it protects is second-tier.

The measure is the layer count at each node transition against Samsung's. Sandisk does not need to lead — it needs the gap not to widen, because in NAND, cost per bit follows density and nothing else.

Moat trajectory: Holding steady

Eight generations shipped on a roughly two-year cadence, with BiCS10 at 332 layers sampling and production expected in 2027. The co-ownership terms are unchanged and the competitive position relative to Samsung is unchanged.

The number that tests this moat

Reported

Research and development expense

$1,328M in fiscal 2026 (6.6% of revenue)

The process technology is developed and owned jointly with Kioxia, so Sandisk's own R&D goes into controllers, firmware and products. R&D falling as a share of a much larger revenue base is fine; falling in dollars would weaken what differentiates it.

Source: Sandisk Form 10-K, FY2026 ↗

Aspects of the moat

DIG DEEPER 01

Eight Generations, and 332 Layers Coming

Eight generations in fifteen years, 332 layers sampling — and in NAND the layer count is not vanity, it is cost per bit and there is nothing else.

DIG DEEPER 02

A Supplier Can Cut You Off; a Co-Owner Cannot

Co-ownership guarantees Sandisk can never be denied the technology, and guarantees a competitor with an identical cost base for as long as it exists.

DIG DEEPER 03

Eight Thousand Patents, and What They Do Not Do

Eight thousand patents, and the filing says out loud that they cannot alone ensure the company's future — because in memory the defensible knowledge…

DIG DEEPER 04

One Roadmap, Two Balance Sheets

R&D is 6.6% of a record year and was 19.2% of a bad one. Splitting that bill with a competitor is what made a NAND roadmap survivable through three…

⚠ Threats to the moat

⚠ THREAT 01

Samsung Is at Four Hundred Layers

Samsung is at 400 layers and Sandisk at 332. That gap is not a scoreboard; it is cost per bit, and there is no marketing answer to it.

⚠ THREAT 02

Shared Technology Is Shared With a Competitor

The technology contest between the third- and fifth-largest NAND producers is a tie by construction, because it is the same technology.

References

ReportedEach also licenses to the other the technologies it developed independently that are reasonably necessary to the joint work. That is a materially different thing from a supply agreement.
Sandisk Corporation, Form 10-K FY2026 — consolidated statements of operations, balance sheets and cash flows, and the results-of-operations and liquidity discussion in Item 7. Revenue net $20,248 million against $7,355 million and $6,663 million in the two prior years, up 175%; cost of revenue $5,776 million (28.5% of revenue); gross profit $14,472 million (71.5%, up 4,100 basis points); research and development $1,328 million (6.6%); selling, general and administrative $676 million (3.3%); loss on debt extinguishment $46 million; business separation costs $25 million; total operating expenses $2,083 million; operating income $12,389 million (61.3%); gain on equity securities $808 million; interest income $70 million; interest expense $73 million; other expense $177 million; income before taxes $13,017 million; income tax expense $1,584 million at a 12% effective rate (against negative 11% and negative 34%); net income $11,433 million (56.5%) against losses of $1,641 million and $672 million. Basic EPS $77.78 and diluted $73.76, on 147 million basic and 155 million diluted weighted average shares. Revenue by end market: Datacenter $5,153 million, $960 million and $325 million; Edge $12,160 million, $4,127 million and $4,069 million; Consumer $2,935 million, $2,268 million and $2,269 million. Revenue by geography: Asia $14,241 million, Americas $4,275 million, EMEA $1,732 million. Datacenter revenue rose 437% with products sold up almost 120% on an exabyte basis and revenue per gigabyte up almost 150%; Edge rose 195% with exabytes up a high single-digit percentage and revenue per gigabyte up almost 180%; Consumer rose 29% with exabytes DOWN a mid-teens percentage and revenue per gigabyte up a low-fifties percentage; total products sold increased by a mid-teens percentage on an exabyte basis. Sales incentive and marketing programmes represented 11%, 19% and 19% of gross revenues in 2026, 2025 and 2024. Balance sheet at 3 July 2026: cash and cash equivalents $4,762 million, accounts receivable $4,708 million, inventories $2,698 million, total current assets $12,780 million, marketable equity securities $1,777 million, property plant and equipment net $674 million, notes receivable and investments in Flash Ventures $678 million, goodwill $4,994 million, total assets $22,507 million; refund liabilities $1,500 million (from $126 million), contract liabilities $849 million current and $393 million non-current, income tax payable $1,286 million, total current liabilities $5,581 million, long-term debt nil (from $1,829 million), total liabilities $6,771 million, treasury stock $4,537 million, retained earnings $9,649 million (from an accumulated deficit of $1,784 million), total shareholders' equity $15,736 million, 149 million shares issued and 146 million outstanding. Cash flows: operating activities provided $11,671 million against $84 million and a use of $309 million; investing used $1,386 million including $970 million of purchases of marketable equity securities, $275 million of net issuances related to Flash Ventures and $177 million of capital expenditures; financing used $7,001 million including $4.5 billion of share repurchases, $1.9 billion of Term Loan repayments and settlement and $630 million of taxes on vested stock awards. Cash conversion cycle 162 days (DSO 48, DIO 178, DPO 64). $2,879 million of cash was held outside the US. Contract liabilities were $1,242 million and refund liabilities $1,500 million under long-term agreements. Unrecognised tax benefits were approximately $323 million. Tax holidays in Malaysia expire at various dates during 2028 through 2031. Total material cash requirements were $11,760 million, of which Flash Ventures-related commitments were $6,559 million and purchase obligations and other commitments $4,902 million. — FY2026 · publ. 2026-08-17 · source ↗
ReportedBiCS10 is 332 layers, built by stacking three 100-plus-layer strings rather than as a monolith, with a 4.8 Gbit/s interface about 33% faster than BiCS8; both partners are sample shipping and mass production is expected in 2027 at Kitakami. For comparison, SK hynix is at 321 layers, Micron at 276, and Samsung at 400.
Coverage of Kioxia and Sandisk's BiCS10 3D NAND. BiCS10 is a 332-layer technology; the interface speed of 4.8 Gbit/s is 33 percent faster than BiCS8, achieved using the Toggle DDR6.0 interface and a Separate Command Address protocol. Mass BiCS10 production is expected to start in 2027 at the Kitakami plant 2 in Iwate Prefecture, Japan, and because Sandisk and Kioxia share the output of the fab through their joint venture both are sample shipping BiCS10 chips; a QLC version of the 332-layer technology would increase chip capacity by a third. The BiCS10 technology involves stacking three 100-plus layer NAND strings together rather than building a monolithic 332-layer chip. Kioxia has an intervening BiCS9 technology which uses BiCS8 218-layer 3D NAND cells with a separate CMOS logic layer providing more performance than the BiCS8 logic circuitry. For comparison, SK hynix has 321 layers in its ninth-generation 3D NAND, also using a triple string stack design; Samsung has 400 layers with its tenth-generation V-NAND, made with separate logic and NAND cell wafers producing a 1 Tbit die in what Samsung calls a Cell-on-Periphery architecture; Micron is at the 276-layer level; and China's YMTC is expected to announce 300-layer class technology. — 2026 · publ. 2026-07-03 · source ↗
ReportedFor comparison, SK hynix is at 321 layers, Micron at 276, and Samsung at 400. Rated narrow.
Coverage of Kioxia and Sandisk's BiCS10 3D NAND. BiCS10 is a 332-layer technology; the interface speed of 4.8 Gbit/s is 33 percent faster than BiCS8, achieved using the Toggle DDR6.0 interface and a Separate Command Address protocol. Mass BiCS10 production is expected to start in 2027 at the Kitakami plant 2 in Iwate Prefecture, Japan, and because Sandisk and Kioxia share the output of the fab through their joint venture both are sample shipping BiCS10 chips; a QLC version of the 332-layer technology would increase chip capacity by a third. The BiCS10 technology involves stacking three 100-plus layer NAND strings together rather than building a monolithic 332-layer chip. Kioxia has an intervening BiCS9 technology which uses BiCS8 218-layer 3D NAND cells with a separate CMOS logic layer providing more performance than the BiCS8 logic circuitry. For comparison, SK hynix has 321 layers in its ninth-generation 3D NAND, also using a triple string stack design; Samsung has 400 layers with its tenth-generation V-NAND, made with separate logic and NAND cell wafers producing a 1 Tbit die in what Samsung calls a Cell-on-Periphery architecture; Micron is at the 276-layer level; and China's YMTC is expected to announce 300-layer class technology. — 2026 · publ. 2026-07-03 · source ↗

Sources

Generated September 23, 2026