✦ The Future BetsNarrow moat

Sandisk (SNDK) — the future bets

One new product that could change what this company is, one roadmap that keeps it alive, $31 billion of somebody else's concrete, and $20 billion of its own stock.

Sandisk's four bets are one new product, one old roadmap, thirty-one billion dollars of somebody else's concrete, and twenty billion dollars of its own stock.

One new product that could change what this company is, and three extensions of what it already does.

High Bandwidth Flash is the only one that could change what this company is. It stacks NAND in an HBM-style package next to an accelerator, claiming 8 to 16 times HBM's capacity at similar cost and bandwidth, and Sandisk taped out the first die ahead of its August 2026 investor day, with samples targeted for 2027.¹ If it works, Sandisk stops being a storage supplier and becomes a memory supplier to AI accelerators. If it does not, nothing else here changes the answer.

BiCS10 is the ordinary version of progress: 332 layers, sample shipping, mass production expected in 2027 at Kitakami.² It is what keeps cost per bit falling, and it is what Sandisk and Kioxia have done every two years for fifteen years.

Thirty-one billion dollars of Japanese capacity is the largest number in the plan and the least under Sandisk's control. Announced on 27 August 2026, over $31 billion — about ¥5 trillion — with Kioxia through 2032, contingent on Japanese government support.³ It is capacity Sandisk will pay for half of and will never own.

Twenty billion dollars of buybacks is the most certain of the four. $6 billion authorised in April, another $14 billion in August, $15.5 billion remaining.⁴ Against a $218.7 billion market value, it is a real return of the windfall.

Rated narrow, trajectory widening — from a thin base. Three of the four are extensions of what Sandisk already does; one is an attempt to build a moat where there is none.

The date that matters is the HBF sample. It has already moved once.

Moat trajectory: Widening

HBF was taped out and specified through the OCP, BiCS10 began sampling, $31 billion of capacity was announced and $20 billion of buybacks authorised — all within twelve months. Three of the four are extensions of what already works.

The number that tests this moat

Reported

GAAP net income, fiscal year

$11,433M in fiscal 2026, from a $1,641M loss

Every bet on these pages is paid for out of peak-cycle profit; the question is what the bets look like when this line turns down.

Source: Sandisk fiscal Q4 2026 results release (Exhibit 99.1, 5 August 2026) ↗

✦ Future bets — beyond today's moat

✦ FUTURE BET 01

High Bandwidth Flash

High Bandwidth Flash is the only thing on these pages that could turn a commodity supplier into a memory supplier — and the sample date has already…

✦ FUTURE BET 02

BiCS10 and the 332-Layer Node

BiCS10 is the bet that requires no new market to exist: 332 layers, sample shipping, and second on a scoreboard Samsung leads.

✦ FUTURE BET 03

Thirty-One Billion Dollars of Japanese Fab

The largest commitment in Sandisk's future is $31 billion of Japanese fab it will pay half of, never own, and cannot approve alone.

✦ FUTURE BET 04

Twenty Billion Dollars of Buybacks

Twenty billion dollars of authorised buybacks, against a company the market valued near six billion eighteen months ago.

References

ReportedIt stacks NAND in an HBM-style package next to an accelerator, claiming 8 to 16 times HBM's capacity at similar cost and bandwidth, and Sandisk taped out the first die ahead of its August 2026 investor day, with samples targeted for 2027. If it works, Sandisk stops being a storage supplier and becomes a memory supplier to AI accelerators.
Coverage of Sandisk's 2026 Investor Day of 13 August 2026 and its High Bandwidth Flash programme. Sandisk has taped out the first High Bandwidth Flash memory die, shown on an HBF Roadmap slide next to a die photograph, with the first HBF inference product samples listed as coming soon in 2027; a tapeout means the design is finished and committed to a mask set, and is several steps short of a product, with wafers still to return from the fab, yields to climb, and the die to survive stacking into 8-high and 16-high configurations with a working logic die and controller, followed by thermal and endurance qualification, accelerator software work and customer qualification cycles. Per Sandisk's HBF fact sheet, the first generation targets 512GB per stack built from sixteen 256Gb die at 1.6 TB/s of read bandwidth, which Sandisk claims lands at up to 8 to 16 times the capacity of HBM at a similar cost, in a package closely matching HBM4's footprint, stack height and power profile, and which is non-volatile and spends no power on refresh; a second generation targets more than 2 TB/s and up to 1TB per stack at 0.8 times the first generation's power, and a third pushes past 3.2 TB/s and up to 1.5TB per stack at 0.64 times the power. The Investor Day deck frames the target workloads as mixture-of-experts LLMs, long context lengths and large KV caches, with the architecture developed using input from major cloud and AI customers, and lays out three deployments: HBF augmenting HBM around an xPU, replacing HBM stacks outright in a similar footprint, or sitting disaggregated holding decode weights and KV cache while a smaller HBM tier acts as cache. Sandisk showed an inference token output comparison from which it draws an 8x capex efficiency claim (one HBF GPU against eight HBM GPUs) and a 2x GPU efficiency claim (four HBF GPUs delivering the same token output as eight HBM GPUs), both labelled as based on internal testing; the fact sheet adds a simulation result of HBF landing within 2.2% of unlimited-capacity HBM when reading pretrained weights for Llama 3.1 405B. These are vendor numbers on unreleased silicon and the chart carries no axis values. On timing, in August 2025 Sandisk said the first HBF samples were targeted for the second half of calendar 2026 with AI inference devices expected in early 2027; as of the August 2026 Investor Day the first HBF inference product samples are listed as coming soon in 2027, and mass production is reported for 2028. On 3 August 2026 Sandisk and SK hynix released the first HBF technical specification through the Open Compute Project, six months after the consortium formed, with Google and Tenstorrent among the contributors. — August 2026 · publ. 2026-08-18 · source ↗
ReportedBiCS10 is the ordinary version of progress: 332 layers, sample shipping, mass production expected in 2027 at Kitakami. It is what keeps cost per bit falling, and it is what Sandisk and Kioxia have done every two years for fifteen years.
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 ↗
ReportedAnnounced on 27 August 2026, over $31 billion — about ¥5 trillion — with Kioxia through 2032, contingent on Japanese government support. It is capacity Sandisk will pay for half of and will never own.
Kioxia and Sandisk, joint announcement of 27 August 2026: the two companies will invest over $31 billion, approximately 5 trillion yen, through 2032 in the ongoing buildout of infrastructure at the Yokkaichi Plant and the Kitakami Plant along with related infrastructure and technology, extending their leadership in the memory industry. The investments are contingent upon government support and are described as aligned with the economic policy goals of the Takaichi administration in supporting a strategically important sector. Kioxia chief executive Hiroo Ota said the joint investment further strengthens the longstanding partnership with Sandisk and underscores Kioxia's commitment to contributing to the advancement of an AI-driven society; Sandisk chief executive David Goeckeler said that for decades Sandisk and Kioxia have jointly developed world-class NAND flash memory technology. — August 2026 · publ. 2026-08-27 · source ↗
Reported$6 billion authorised in April, another $14 billion in August, $15.5 billion remaining. Against a $218.7 billion market value, it is a real return of the windfall.
Sandisk Corporation (Nasdaq: SNDK) market data — share price about $1,495, market capitalisation about $218.7 billion on approximately 146.4 million shares outstanding, trailing price/earnings about 20.4 and forward price/earnings in the single digits, on trailing revenue of $20.25 billion and trailing net income of $11.43 billion with trailing EPS of $73.76; no dividend. The all-time low was $27.89 on 7 April 2025 and the all-time high $2,354.39 on 22 June 2026, with a 52-week range of $48.56 to $2,354.39. Consensus from 19 analysts is for fiscal 2027 revenue of about $41 billion and earnings per share of about $177, recently raised from $33 billion and $112. — August 2026 · publ. 2026-08-28 · source ↗

Sources

Generated September 23, 2026