⚠ Sharing a Roadmap Removes the Option of a Different OneModerate threat

Sandisk (SNDK) — threat to the moat

Sandisk built its most consequential future product with SK hynix rather than with the partner it shares a roadmap with. That is worth noticing.

Sharing the roadmap removes the option of taking a different one.

One roadmap, and the options it removesPursue a different architecture in FVRequires agreementSkip a nodeNoAccelerate a nodeNoMinimum shared R&D, 2027$138MHBF specification authored withSK hynix, not KioxiaPublished throughthe Open Compute ProjectR&D was 19.2% of revenue in FY2023 and Sandisk still lost $2.1bn. Sharing it made it survivable.
Sandisk built its most consequential future product with a different partner, which is worth noticing.

Sandisk cannot pursue an architecture inside Flash Ventures that Kioxia does not want, cannot skip a node, cannot accelerate one, and cannot fund a divergent programme with the ventures' equipment. The technologies are co-developed for Flash Ventures' use and jointly owned, and Sandisk is committed to a minimum level of common R&D funding — $138 million for 2027.1

For fifteen years that has cost nothing, because there has been one obvious direction: more layers, lower cost per bit. It starts to matter as the interesting questions in memory stop being about density. High Bandwidth Flash, 3D matrix architectures, storage-class memory — these are places where two companies with different customer bases might reasonably reach different conclusions about what to build and when.

Note where Sandisk went for its most important new bet. The HBF specification was authored with SK hynix, not with Kioxia, and published through the Open Compute Project.2 That is a sensible way to make a standard, and it is also Sandisk building its most consequential future product with a different partner.

The offsetting fact is the one that justified the arrangement in the first place: R&D was 19.2% of revenue in fiscal 2023 and the company still lost $2.1 billion.3 Splitting the bill is what made the roadmap survivable.

Watch whether HBF is manufactured inside Flash Ventures or outside it. The answer determines whether the joint venture is Sandisk's future as well as its past.

References
  1. ReportedThe technologies are co-developed for Flash Ventures' use and jointly owned, and Sandisk is committed to a minimum level of common R&D funding — $138 million for 2027. For fifteen years that has cost nothing, because there has been one obvious direction: more layers, lower cost per bit.
    Sandisk Corporation, Form 10-K FY2026 — Note 10, Related Parties and Related Commitments and Contingencies, and the related Item 1 and Item 7 disclosure on Flash Ventures. Sandisk procures all of its flash-based memory wafers from Flash Partners Ltd., Flash Alliance Ltd. and Flash Forward Ltd.; it holds a 49.9% ownership interest and Kioxia 50.1% in each. Wafers are manufactured by Kioxia at its wafer fabrication facilities in Japan using equipment individually owned or leased by each Flash Ventures entity; each entity purchases wafers from Kioxia at cost and resells them to Sandisk and Kioxia at cost plus a markup, with each partner generally entitled to 50% of output. The facilities are Y3 (Flash Partners, 2004), Y4 (Flash Alliance, 2006), Y5 (Flash Forward, 2010), New Y2 (production from 2016), Y6 (2018), K1 at Kitakami (2019), Y7 (2022) and K2 at Kitakami, whose output began in the year ended 3 July 2026. Sandisk is obligated to pay for variable costs based on a rolling three-month forecast, and purchase orders placed with Flash Ventures for up to three months are binding and cannot be cancelled; it is obligated to pay for half of Flash Ventures' fixed costs regardless of the output it chooses to purchase, and is committed to fund 49.9% to 50.0% of each entity's capital investments where operating cash flow is insufficient. Flash Ventures has historically operated at approximately 100% of manufacturing capacity; during 2026, 2025 and 2024 Sandisk temporarily reduced its utilisation and incurred costs of $11 million, $75 million and $249 million respectively, recorded as charges to cost of revenue. It participates in common R&D activities with Kioxia and is contractually committed to a minimum funding level, with R&D commitments due for 2027 of $138 million. Sandisk accounts for its ownership under the equity method; the entities are VIEs and Sandisk determined it is not the primary beneficiary of any of them, concluding on its 49.9% ownership, the voting structure and the manner in which day-to-day operations are conducted that it lacked the power to direct most of the activities that most significantly impact economic performance. Its 49.9% interest in the earnings of the entities is recognised one quarter in arrears in Other income (expense), net. Summarised financial information for the ventures: net sales of $2,775 million, $2,315 million and $2,252 million and net losses of $85 million, $63 million and $9 million for 2026, 2025 and 2024, with a gross loss of $93 million in 2026; total assets $7,320 million and total net equity of investees $1,025 million. Notes receivable and investments in Flash Ventures totalled $679 million. Sandisk made net payments to Flash Ventures of $3.6 billion, $3.4 billion and $3.4 billion in 2026, 2025 and 2024 and received distributions of $107 million and $176 million in 2026 and 2025. Maximum estimable loss exposure is $2,897 million: notes receivable $577 million, equity investments $102 million, operating lease guarantees $923 million and inventory and prepayments $1,295 million. Flash Ventures sells to and leases back a portion of its tools from a consortium of financial institutions; Sandisk guarantees half of all outstanding obligations under each lease agreement, totalling JPY149.0 billion or $923 million, in annual guarantee instalments of $378 million in 2027, $245 million in 2028, $126 million in 2029, $96 million in 2030 and $78 million in 2031. Prepayments toward Sandisk's share of future building depreciation of $840 million remain to be credited against future wafer purchases, with a further $402 million committed through fiscal 2035. On 29 January 2026 the FAL and FPL Second Commitment and Extension Agreements extended Flash Alliance and Flash Partners from 31 December 2029 to 31 December 2034, so that all three ventures co-terminate on that date; an Agreement to Enhance Collaboration commits Sandisk Technologies to pay Kioxia $1.2 billion over 2026 through 2029 in consideration of Kioxia's manufacturing services and the continued availability of supply, amortised straight-line into cost of revenue. On expiry each entity commences a wind-up and is dissolved, with net proceeds distributed in kind or cash pro rata. Flash Ventures-related commitments total $6,559 million: $2,627 million in 2027, $2,577 million in 2028-2029, $1,318 million in 2030-2031 and $37 million beyond. — FY2026 · publ. 2026-08-17 · source ↗
  2. ReportedThe HBF specification was authored with SK hynix, not with Kioxia, and published through the Open Compute Project. That is a sensible way to make a standard, and it is also Sandisk building its most consequential future product with a different partner.
    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 ↗
  3. ReportedThe offsetting fact is the one that justified the arrangement in the first place: R&D was 19.2% of revenue in fiscal 2023 and the company still lost $2.1 billion. Splitting the bill is what made the roadmap survivable.
    Sandisk Corporation, Form 10-K for the fiscal year ended 27 June 2025 (SEC, CIK 2023554) — consolidated statements of operations for fiscal 2025, 2024 and 2023. Revenue net $7,355 million, $6,663 million and $6,086 million; cost of revenue $5,143 million, $5,591 million and $5,656 million; gross profit $2,212 million (30.1%), $1,072 million (16.1%) and $430 million (7.1%); research and development $1,132 million (15.4%), $1,061 million (15.9%) and $1,167 million (19.2%); selling, general and administrative $573 million, $455 million and $558 million; goodwill impairment $1,830 million in 2025 and $671 million in 2023; total operating expenses $3,589 million, $1,540 million and $2,465 million; operating loss $1,377 million, $468 million and $2,035 million; net loss $1,641 million, $672 million and $2,143 million. Net loss per common share, basic and diluted, of $11.32, $4.63 and $14.78 on 145 million weighted average shares in each year. Revenue by end market on the prior labels: Cloud $960 million, $325 million and $500 million; Client $4,127 million, $4,069 million and $3,637 million; Consumer $2,268 million, $2,269 million and $1,949 million. Revenue by geography: Asia $4,457 million, $4,510 million and $3,890 million; Americas $1,618 million, $1,095 million and $1,266 million; Europe, Middle East and Africa $1,280 million, $1,058 million and $930 million. No customer accounted for more than 10% of net revenue in any of the three years. — FY2025 · publ. 2025-08-21 · source ↗
Sources
Generated September 23, 2026