SanDisk Stock (SNDK) Q4 Earnings Preview: The HBF Bet

The datacenter line is the one that matters strategically. Revenue from data center customers rose 645% year over year and 233% sequentially, which means SanDisk is no longer primarily a consumer flash company. That mix shift toward enterprise SSDs is what drove the margin expansion, because negotiated enterprise contracts carry far better economics than commodity retail flash.

Investment takeaway: A 78.4% gross margin in a business historically defined by commodity pricing signals genuine scarcity, not clever management. The question is how long scarcity lasts.

What Wall Street Expects From SanDisk’s Q4 Earnings

The fiscal fourth-quarter print lands after the close today, August 5, and expectations sit above management’s own numbers.

SanDisk guided revenue of $7.75 billion to $8.25 billion with non-GAAP earnings of $30.00 to $33.00 per share. Analyst consensus has drifted toward roughly $8.3 billion in revenue — above the top of the company’s guided range — with EPS estimates clustered near $34, also above the guide. That gap is meaningful: SanDisk stock is priced for a beat, not merely a hit.

Goldman’s James Schneider raised his target from $1,200 to $2,200 on July 5, built on a 20-times multiple applied to a normalized earnings estimate near $110 per share; his calendar-2026 earnings estimate sits more than 30% above the Street. Bernstein went further, lifting its target to $3,000 from $1,700 while keeping an Outperform rating, arguing that SanDisk’s multi-year supply agreements — with a price floor near $0.29 per gigabyte — offer more downside protection than earlier take-or-pay contracts. The Street-high target sits near $3,050, while the average has been far lower and highly volatile through the summer, a spread that captures how unsettled the debate remains.

A second catalyst follows quickly. SanDisk holds an investor day on August 13, eight days after earnings, featuring CEO David Goeckeler and CFO Luis Visoso, where management is expected to quantify its long-term agreement pipeline.

Investment takeaway: With consensus above guidance, a merely in-line quarter could disappoint. The August 13 investor day may matter more than the print itself for the multi-year thesis.

What Management Is Signaling

Management has framed the quarter as a structural turn rather than a cyclical spike. CEO David Goeckeler called the period a fundamental inflection point, pointing to a deliberate mix shift toward the highest-value end markets — led by datacenter — and a new business model built on multi-year customer engagements backed by firm financial commitments. Alongside the Q3 results, the company also authorized a $6 billion share buyback, a signal of balance-sheet confidence even as it funds capacity expansion.

Investment takeaway: The buyback and the contract language point in the same way — management is trying to convince investors that this cycle’s earnings are more durable than NAND’s history would suggest.

HBF: SanDisk Builds a New AI Memory Tier

The technology story centers on a category SanDisk is trying to create rather than compete within.

On August 3, SanDisk and SK hynix released the first High Bandwidth Flash technical specification through the Open Compute Project, six months after forming the workstream in February. Publishing through OCP makes it an open industry standard rather than a proprietary format, a deliberate choice to become the default before rivals define one.

The specification also defines electrical characteristics, reliability, and packaging guidelines for the die-stack process, and software input-output guidance. Google and Tenstorrent joined during standardization and contributed to technology validation, which matters because Google is exactly the kind of hyperscale customer that would have to adopt HBF for it to succeed.

Investment takeaway: Publishing openly through OCP trades short-term exclusivity for the chance to set the industry default. If HBF becomes standard, SanDisk sells into every implementation rather than defending a closed format.

Why Inference, Not Training

The strategic insight behind HBF is a deliberate decision about which battle to fight.

High-bandwidth memory, built on DRAM, dominates AI training and will continue to. HBF targets a different problem: inference workloads where models are too large to fit in HBM but need far more speed than a conventional SSD delivers. SanDisk has claimed HBF can provide eight to sixteen times more capacity than HBM at a comparable cost per package.

The performance evidence is a simulation rather than silicon, and should be read as such. Running Meta’s Llama 3.1 405-billion-parameter model, SanDisk reported HBF landing within 2.2% of a hypothetical system with unlimited HBM capacity. Crucially, SanDisk and SK hynix are not positioning HBF as an HBM replacement, but as a tier that coexists with it — HBM handling latency-critical work while HBF holds model weights and inference data.

Chief Technology Officer Alper Ilkbahar has framed the distinction around access patterns. Inference reads data repeatedly but rarely rewrites it, which plays directly to NAND’s strengths and away from its weaknesses.

Investment takeaway: Choosing inference over training is a defensible niche strategy rather than a head-on fight with Samsung and SK hynix in HBM. It also means HBF’s fate depends on inference workloads growing faster than HBM capacity does.

BiCS10: Winning Without the Layer Race

The second technology pillar is a rebuke to how the NAND industry usually keeps score.

BiCS10, developed with Kioxia, stacks 332 layers and reaches a bit density above 29 gigabits per square millimeter on a 1-terabit TLC device. That is a 59% density improvement over the 218-layer BiCS8 generation. Interface speed rises to 4.8 gigabits per second, 33% faster than the flash currently in mass production, which supports PCIe Gen5 and Gen6 data center drives.

The competitive point is that Samsung has announced NAND exceeding 400 layers, yet BiCS10 achieves higher areal density with fewer layers, because lateral scaling and floor-plan efficiency matter as much as stack height. Fewer layers also simplify vertical etching and reduce wafer warpage, which lowers cost.

The architecture uses CMOS directly bonded to the array, fabricating logic and memory-cell wafers separately, then bonding them, paired with On-Pitch Select Gate Drain technology that shortens bit lines and cuts capacitance. The two partners describe the efficiency gains differently: SanDisk cites a 10% reduction in data input power and 34% in data output power, while Kioxia frames it as 18% and 30% improvements in write and read power efficiency. A QLC variant, formally unveiled at FMS 2026, reaches roughly 37 gigabits per square millimeter — described by the partners as the densest 3D NAND publicly introduced to date.

Investment takeaway: Density per square millimeter, not layer count, determines cost per bit. BiCS10 gives SanDisk a genuine cost advantage in exactly the data center segment where pricing is strongest.

The Kioxia Alliance and Japan’s Revival

SanDisk’s manufacturing runs through a decades-old joint venture, and that partnership has become a geopolitical story.

SanDisk and Kioxia jointly operate fabs at Yokkaichi and Kitakami in Japan, sharing output. The alliance raised capital investment for 2026 by roughly 40% year over year to about $4.5 billion, expanding current capacity while building BiCS10 capability at Kitakami Fab 2 in Iwate Prefecture. Mass production of BiCS10 is expected in 2027.

The scale of Japan’s memory revival is best captured by Kioxia itself. Its shares rose roughly sevenfold on the AI memory boom, and on June 12 its market capitalization briefly surpassed Toyota Motor at above $270 billion, making it Japan’s most valuable listed company — before giving back much of that gain in the same July correction that hit SanDisk. For a country whose semiconductor share fell from roughly half the global market in the 1980s to under 10% in recent decades, that is a striking reversal.

SanDisk’s own structure reinforces the theme. It separated from Western Digital in February 2025 to operate as an independent, US-headquartered NAND specialist competing against Samsung, SK hynix, and China’s YMTC.

Investment takeaway: The Kioxia joint venture gives SanDisk scale it could not fund alone, but it also means capacity decisions are shared rather than controlled. That is a structural constraint in a market where supply discipline drives pricing.

The Business Model Shift

Perhaps the most consequential change is commercial rather than technical.

NAND has historically swung between painful oversupply and extreme scarcity, with pricing set largely on the spot market. SanDisk signed five New Business Model agreements during and shortly after its third quarter — arrangements carrying firm financial commitments designed to make revenue durable through a downturn. A growing share of future bit output is already committed under these agreements (specific coverage figures should be re-verified against the Q4 release).

The industry is moving in the same direction. Micron rose sharply in late July after Samsung indicated memory shortages could persist through 2027 and 2028, suggesting tightness extends beyond any single company’s execution.

Investment takeaway: Contracted multi-year pricing is the single most important structural change in this cycle. If it holds through the next downturn, NAND deserves a higher multiple than history suggests. If it breaks, nothing else in the thesis matters.

SanDisk Stock After a Sharp Summer Correction

Here is where the enthusiasm meets the tape.

SanDisk stock rose more than 700% year-to-date at its peak, briefly the best performer in the S&P 500. It reached a 52-week intraday high of $2,354.39 on June 22, closed July 9 at $1,858.27, and then corrected hard alongside the entire memory and storage complex — Seagate, Micron, and Western Digital all slid with it — closing August 4 around $1,427.62, roughly 39% below the June peak.

The bull case is that the correction was sentiment, not fundamentals, and that a company compounding revenue at 251% with 78% gross margins is not expensive on forward earnings. Goldman’s normalized $110 earnings estimate implies SanDisk stock trades at a modest multiple if that number proves achievable.

The bear case is that a drawdown this steep, this fast, reveals how much of the move was momentum, that consensus now sits above company guidance, and that real uncertainty remains over which storage architecture wins the inference era. HBF samples do not arrive until the second half of 2026, and commercial adoption is later still.

Investment takeaway: SNDK suits investors who believe contracted pricing has structurally changed NAND economics and can tolerate 40%-plus drawdowns. It does not suit anyone needing stability in an earnings print where expectations exceed guidance.

The Risks That Matter for SanDisk Stock

Expectations above guidance. Consensus near $8.3 billion sits above management’s $7.75 to $8.25 billion range, so an in-line quarter could disappoint.

Extreme volatility. A roughly 39% pullback from the June high, arriving within weeks of a 700%-plus run, demonstrates how violently this stock moves on sentiment.

Cyclicality. NAND’s pricing power has collapsed in every prior cycle once supply caught demand, and contracted agreements are untested through a real downturn.

HBF is unproven. The performance evidence is a simulation; samples arrive in the second half of 2026, and commercial adoption comes later.

Architecture risk. It is still unsettled which storage architecture the AI inference era ultimately adopts, and HBF is one bet among several.

Shared capacity control. The Kioxia joint venture means supply decisions are negotiated, not unilateral.

Competitive response. Samsung, SK hynix, and YMTC are all scaling, and Samsung already has a 400-plus-layer NAND announced.

Closing Thoughts on SanDisk Stock

SanDisk enters tonight’s earnings with a genuinely differentiated position. BiCS10 delivers leading areal density without chasing the layer count. HBF has become a published open standard with Google and Tenstorrent involved, and contracted agreements now cover a growing share of next year’s bit output. Independence from Western Digital has been converted into technical credibility faster than most spinoffs manage.

The verdict rests on one question: can new multi-year contract structures outlast the NAND industry’s historically brutal boom-and-bust cycle? Watch four things beyond the headline numbers: fiscal 2027 pricing and margin guidance, commentary on how quickly HBF samples convert into paying customers, whether the New Business Model contracts hold if spot pricing softens, and what management commits to at the August 13 investor day. The technology is real. The summer drawdown was a reminder that the price is a separate question entirely.