Sandisk's 85% Gross Margin Meets a 43% Drawdown

The NAND upcycle is real. The stock is already arguing about its duration.

SNDK · Earnings Call · 2026-10-29

The quarter that should have ended the argument

Sandisk just printed the kind of quarter that usually ends a bear debate rather than starts one. “Revenue for the quarter was $11 billion. Non-GAAP gross margin was 85.1% and non-GAAP earnings per share was $49.51.” — David Goeckeler, Chairman and Chief Executive Officer · 2026-10-29 It generated $5.7 billion of adjusted free cash flow, returned $5 billion through buybacks, closed with $5.1 billion of cash and no debt, and announced an 8-for-1 stock split. That is not a cyclical rebound. That is a different business model wearing the same ticker. The tape disagrees. Sandisk shares are down 30.8% over the last 90 trading days and 43% below the June 25 peak of $2,335. The market cap is still $192 billion. So the income statement is going up and to the right while the stock is going down and to the left. The market is not disputing the quarter; it is disputing the durability of 85% gross margins. The last filed quarter already showed the transformation: revenue of $9.0 billion, up 372% year over year and 51% sequentially, with gross margin at 84.6% and free cash flow of $7.0 billion. The just-reported quarter took revenue to $11 billion and gross margin to 85.1%. Those numbers are so far outside Sandisk's historical range that the central investment question is no longer demand. It is whether this is a new baseline or a spectacular peak.

The narrative rotated from contracts to architecture

The most interesting shift on this call is not in the numbers. It is in the language. In prior quarters, management sold visibility: the long term agreement structure, financial guarantees, and the remaining performance obligation. In April, Luis Visoso anchored the story with “about $42 billion of RPO in this business.” — Luis Visoso, Likely Chief Financial Officer or Senior Finance Executive · 2026-04-30 In August, David Goeckeler described “more than 50% of our supply for fiscal year '27 ... In FY '28, that steps up to 2/3” — David V. Goeckeler, CEO · 2026-08-05 of supply committed. This quarter, those same contract concepts are present but no longer lead the keyword set. The fresh high-momentum themes are inference architecture, KV Cache, BiCS 9, and customer partnership. The old financial guarantees and long term agreement language now reads as background. Notably, the transcript did not refresh the $42 billion RPO figure that had anchored prior calls. Instead, management talked about a pipeline: Goeckeler said the NBM pipeline exceeds supply through the end of the decade and that customers are coming back for second tranches. That is a subtle but important rotation. The bull case needed the NBM structure to de-cycle NAND and turn memory into something closer to contracted infrastructure. This call instead asked investors to underwrite an architecture cycle — financial model and all — while the stock was already de-rating the old contract story. The contracts are still working; they are just no longer the new thing.

in general, we’re hearing from our customers that they have underestimated the amount of NAND they need to run their infrastructure.
David Goeckeler, Chairman and Chief Executive Officer · 2026-10-29

The inference architecture is the new demand curve

The operational case is concentrated in AI inference. Goeckeler explained that every generated token must reference prior context, creating a growing KV cache that behaves like the model's working memory. Longer prompts, complex reasoning, multimodal inputs, and persistent user context all expand storage requirements. TLC handles lower-latency, higher-endurance workloads; QLC handles density; and the emerging agent orchestration layer multiplies the number of concurrent sessions that need persistent context. The result is visible in the mix. “Data center revenue was $4.8 billion, up 61% sequentially, outpacing estimated market growth for the quarter.” — Luis Visoso, Executive Vice President and Chief Financial Officer · 2026-10-29 Management estimates data center has grown from roughly 20% of industry bits early in the decade to about half today. Sandisk is still catching up to that mix, which is why data center is the engine while Edge grows more slowly and consumer revenue falls as higher prices reset demand. The technology roadmap supports the pitch. BiCS 9 pairs the proven BiCS 8 cell array with next-generation CMOS for faster interfaces, while BiCS 10 is further out. High-Bandwidth Flash reached tape-out of its first memory die, with samples of HBF-based products targeted for next calendar year — but HBF is explicitly not in the long-term model. That is honest, and it also means the optionality is real but unearned. The most provocative technical point was SLC. Customers are exploring SLC NAND in parts of the architecture where endurance matters more than density, potentially substituting NAND for DRAM. Goeckeler sized the incentive: “the cost differential is huge, 20x to 40x.” — David Goeckeler, Chairman and Chief Executive Officer · 2026-10-29 If that happens at scale, it expands the NAND TAM rather than cannibalizing TLC or QLC. It also consumes far more bits per unit of storage, which is why even a little SLC could tighten the market.

Why the tape is skeptical

The guidance was strong: fiscal Q2 revenue of $12.2 billion to $12.6 billion, gross margin of 84.5% to 85.5%, and pre-split EPS of $55 to $57. But the same call contained the reason the stock cannot hold a premium multiple. Luis Visoso warned that fiscal Q3 is normally the weakest sequential quarter and that bits had fallen 19% sequentially in the comparable period. “Q3 is normally our weakest sequential quarter ... last quarter, for example, our bits were down 19% sequentially.” — Luis Visoso, Executive Vice President and Chief Financial Officer · 2026-10-29 In a stock priced for durable compounding, a seasonal air pocket matters. There is also the supply response. Capital spending is expected to rise to a mid-single-digit percentage of revenue. Nodal transitions remain the primary bit-growth lever — nodal transition is now a cooler theme than it was earlier in the upcycle — but the market has started to ask what happens when pricing stops doing the work. Sandisk's NBM structure limits spot upside even as it protects downside. That is a feature of the financial model, but it also caps the explosive earnings revisions that momentum investors pay for. The tape is making the same point across the complex. The global AI data center theme is down 5.3% over 90 days with 49 decliners against only 7 advancers, and high-bandwidth memory is down 6.9% with all 11 tracked names negative. Sandisk is not idiosyncratically broken. It is the high-beta expression of an AI infrastructure trade that is being marked down while the fundamental data is still accelerating.

The capital return is real, but not the debate

Sandisk repurchased $5 billion of stock in the quarter and has retired about 4% of shares over two quarters. Management reiterated that 100% of excess cash goes back to shareholders, and the 8-for-1 split improves accessibility without changing value. That is shareholder-friendly, and it is a signal of confidence. But at a $192 billion market cap, a $5 billion quarterly buyback is supportive, not transformative. The stock will not re-rate on capital return alone. What would change the debate is evidence that 85% gross margin is structural rather than scarcity-driven: a refreshed RPO, a fiscal Q3 that does not seasonally roll over, and customer commitments that extend beyond the current AI capex cycle. Until then, Sandisk has the income statement of a boom and the stock chart of a bust. The quarter proved the upcycle. The tape is already voting on what comes after it.