SanDisk: The Complete Guide

The fourth quarter alone shows the slope: "Fiscal fourth quarter revenue was $8.97 billion, up 51% sequentially", with "GAAP net income reported at $6.90 billion ($43.97 diluted net income per share)" and a gross margin of 84.6% 3. The shift in what SanDisk sells is as striking as the total — it renamed its end markets between years, and on the new basis FY2026 Datacenter revenue was $5,153 million against $960 million in FY2025, which SanDisk describes as "Datacenter up 437%" 3. Alongside the results it announced "an additional $14 billion buyback program, bringing total remaining authorization to $15.5 billion" 3.

One legal detail shapes how the rest of this page reads. The S-1 states that "Sandisk Corporation was incorporated in Delaware for the purpose of holding WDC's Flash Business in connection with the separation" 9. This is a new registrant, with its own SEC identity, distinct from the SanDisk Corporation that Western Digital bought in 2016 10. Its filings do not claim the older company's founding date; the FY2025 annual report goes no further than "We have a rich heritage of innovation and operational excellence" 2. See The story, below, for what is and is not documented about that heritage.

Two dates circulate for the separation and both are right, which is worth stating rather than choosing between. The legal distribution was 21 February 2025 — the annual report says "As of February 21, 2025, we separated from WDC" 2, and Western Digital's 8-K times it to "February 21, 2025 at 11:59 p.m. Pacific time" against a record date of 12 February 13. The public completion and Nasdaq listing were announced on 24 February, which is the date SanDisk's own FAQ uses: "On February 24th, 2025, Sandisk officially completed its separation from Western Digital as an independent, publicly traded company" 17 18.

On the pre-2016 heritage, this page is deliberately careful. The 1988 incorporation is documented in the old company's own filings 10, which also record that it had "successfully developed and commercialized 11 generations of NAND" 10. But no current SanDisk page found in this check names the founders or claims the 1988 founding as its own, so this page does not attribute either to today's company.

On the flash itself, SanDisk and Kioxia ship the same silicon under different names, because they jointly develop and jointly own the output. BiCS10 is SanDisk's label for the generation Kioxia calls its tenth: "332 memory layers", die density "greater than 29Gb/mm²", "Up to 4.8Gb/s NAND interface speed, a 33 percent improvement", and bit density improved "by 59 percent" over the eighth generation 19. The QLC variant reaches ">37 Gb/mm²" 20. Both are built with "CMOS directly Bonded to Array (CBA) technology", where logic and memory arrays are fabricated on separate wafers and then bonded 19.

One node in the range is a hybrid worth noting: SanDisk describes a BiCS9 QLC part combining "a proven BiCS8 array with a BiCS10-based CMOS wafer" 26 — possible precisely because CBA separates the two wafers, so the generations can be mixed.

The manufacturing base grew by one fab between SanDisk's first two annual reports, which is a clean way to see the joint venture expanding. FY2025 described "seven flash-based manufacturing facilities in Japan, six of which are located in Yokkaichi, Japan and one of which is located in Kitakami, Japan", and said an eighth would begin operations in calendar 2025 4; FY2026 describes "eight flash-based manufacturing facilities in Japan, six of which are located in Yokkaichi, Japan and two of which are located in Kitakami, Japan" 1. The eighth is Kitakami Fab2, which began operation in September 2025 to produce 218-layer BiCS8 with CBA, with "meaningful output expected to begin in the first half of 2026" 21.

SanDisk's own facilities are back-end only. Its 10-K describes assembly and test as comprising "in-house assembly and test facilities located in Penang, Malaysia, facilities operated by other contract manufacturers, and the assembly and test facility owned and operated by SDSS" 1 — and even that last one is mostly someone else's, being "a venture that is owned 20% by Sandisk and 80% by JCET Management Co., Ltd." 1.

The forward commitment is joint too, and large. In August 2026 SanDisk and Kioxia announced "Over $31 billion (approximately 5 trillion yen)" of investment through 2032 across Yokkaichi and Kitakami, noting the partnership has "invested over $50 billion in Japan over the past 25 years" and that the new money is "contingent upon government support" 29. No split between the two companies is disclosed 29. Goeckeler ties it to demand rather than ambition: "In line with our business strategy and financial guidance, these planned investments will ensure our ability to support our customer's increasing demands for our technology" 29.

This check found no SanDisk-published figure for wafer starts per month or for its share of industry bit output; the capital commitment is the only forward capacity number the company publishes 29.

The most interesting thing SanDisk is building is not a denser drive but a different place for flash to sit. High Bandwidth Flash puts NAND in the position HBM occupies — stacked beside the processor — on the argument that AI inference needs capacity more than it needs DRAM's write endurance. SanDisk's own technical write-up describes "512GB total capacity per 16-die stack", matching "HBM4's physical footprint, power profile, and stack height", built on "CBA technology (CMOS directly Bonded to Array)" with "ultra-low warpage designs" 30. Its headline claim is bandwidth parity at far greater capacity: HBF is designed to deliver "comparable bandwidth to High Bandwidth Memory (HBM) while delivering up to 8-16x the capacity of HBM at a similar cost" 25. See The HBM position for where that stands.

The HBF specification is the most detailed public document SanDisk is party to, and it is worth reading as a standard rather than a press release. Published through the Open Compute Project as the "High Bandwidth Flash (HBF) High-Level Base Die Specification", v0.7.0, it defines a maximum of 16 NAND dies per cube for "512 GiB" total, and three performance grades — "0.384 TB/s" at 8 GT/s, "1.536 TB/s" at 16 GT/s and "3.072 TB/s" at 32 GT/s, all on a x64 interface, over "Universal Chiplet Interconnect Express (UCIe) architecture" 32. It describes HBF as "a non-coherent memory-centric flash device" addressing "the memory wall problem", delivering "8x to 16x of the capacity of HBM, while delivering the same read bandwidth" 32. SK hynix's own account of the same specification gives the range as "approximately 0.4TB/s to 3.0TB/s" and positions HBF as "a new memory layer between HBM and SSDs" 33.

The one internal price SanDisk does disclose is the one it pays. Wafers from the joint venture transfer at "cost plus a small markup" 1 — an unusually direct statement of what its own supply costs, and a reminder that SanDisk's margin is made downstream of the wafer, in product design, packaging and mix. That mix shift is what its Investor Day targets rest on: a FY2028–FY2030 model of "approximately 80 percent" non-GAAP gross margin and "approximately 75 percent" non-GAAP operating margin 26.

What the specification actually covers is broader than a memory part: "system interface, electrical and other technical guidelines for designing systems that interact with and use HBF technology, including basic performance expectations, the xPU-HBF host interface, reliability and packaging guidance for an HBF die stack, and a software user guide for read and write operations", explicitly enabling HBF "to coexist with High Bandwidth Memory" 15. That last phrase is the strategic claim — HBF is pitched as an additional tier beside HBM, not a replacement for it, which is also how SK hynix describes it: "a new memory layer between HBM and SSDs" 33.

The evidence SanDisk offers is a simulation rather than a measurement, and the distinction matters: its technical blog states that "HBF's simulated performance on a Llama 3.1 405B parameter model for reading pretrained weights is within a 2.2% delta of a hypothetically unlimited-capacity HBM" 30. Both hedges are Sandisk's own — simulated, and against a hypothetical part rather than a real one. As of this check the specification exists and samples were targeted for the second half of 2026 25; no shipping HBF product is announced. Note also that HBF is a SanDisk and SK hynix effort — Kioxia, SanDisk's manufacturing partner in everything else, is not a party to it 5.

SanDisk is also changing how it sells, not just what. At its 2026 Investor Day it reported that eight customers had signed what it calls New Business Model agreements, covering roughly half of FY2027 bits and about two-thirds of FY2028 bits 26 — long-term commitments that trade spot upside for visibility, the same move Kioxia and Micron have each described in their own terms. Its stated market view is that enterprise datacentre flash demand grows "to 1.2 zettabytes by 2030" 26. That is SanDisk's own projection, presented to investors, and this page reports it as such.

The second thing hidden in plain sight is the name. Today's Sandisk Corporation is a Delaware company "incorporated…for the purpose of holding WDC's Flash Business in connection with the separation" 9, with a different SEC registrant identity from the SanDisk that existed until 2016 10. Meanwhile the products carry a tangle of borrowed brands: WD_BLACK flash drives are SanDisk's, made "under license from Western Digital", while WD_BLACK hard drives are Western Digital's 17. SanDisk began untangling this in January 2026 by retiring the WD names from its own NVMe SSD lines in favour of SANDISK Optimus 24. A shopper comparing two WD_BLACK drives in 2025 was, without any indication on the box, choosing between two different companies' products.

There is also a timing risk specific to its biggest idea. HBF's value depends on an ecosystem — processors that can address it, software that can use it, and a standard that others adopt. SanDisk has done the standards work properly, through OCP and with SK hynix, Google and Tenstorrent involved 15, but as of this check the specification is at v0.7.0 32 and no shipping product exists. The stated milestones, samples in H2 2026 and devices in early 2027 25, are the test.

A narrower disclosure point: SanDisk publishes no wafer-capacity or bit-share figure of its own, so its scale can only be inferred from revenue and from the joint venture's fab count 1 29.

This check found no official SanDisk statement of any Indian manufacturing, packaging or assembly-and-test venture, in any press release, 10-K or registration statement 1 9. Where reports of such ventures circulate, none is corroborated by a SanDisk document found here, so none is repeated as fact. Among the memory makers covered in this section, Micron remains the only one with a plant in India.

The near-term guidance is concrete and dated: Q1 FY2027 revenue of $10.30 billion to $10.80 billion, GAAP gross margin of 83.0% to 84.9%, and non-GAAP diluted EPS of $44.00 to $46.00 3. Whether that held is knowable after this check and was not knowable at it.

The open question is HBF. SanDisk has a published standard, two named consortium partners beyond SK hynix, and a stated sample date 15 25 — but no product and no announced customer. If HBF works, SanDisk becomes a supplier into AI accelerator packages rather than only into storage bays, which would be the largest change in what this company is since the split. If it does not, it remains a flash maker without a fab in a market where its partner decides the capacity.