SanDisk (Trans): NBM pipeline oversupplied, announces 8-for-1 split

Below is Dolphin Research's compiled Trans of SanDisk's FY27 Q1 earnings call.

I. Core takeaways from the print

1. Shareholder returns and stock split

a. Repurchased ~3.2 mn shares this quarter, spending $5.0 bn. Over the past two quarters, cumulative buybacks reached 6.0 mn shares, or ~4% of total shares outstanding, and the company will continue to return 100% of excess cash. b. The BOD approved an 8-for-1 stock split. New shares are expected to be distributed on Nov 19, and the stock will trade on a split-adjusted basis starting Nov 20.

2. FY27 Q2 guidance

a. Revenue of $12.2–12.6 bn. b. Non-GAAP GPM of 84.5%–85.5%. This includes a modest like-for-like price increase. c. Non-GAAP EPS: $55–57 pre-split (diluted shares 151 mn). Post-split, $6.88–7.13 (diluted shares ~1.208 bn).

3. Key metrics this quarter

a. Revenue of $11.0 bn, +23% QoQ and +378% YoY, above the $10.3–10.8 bn guide. About 60% of the upside came from volume and ~40% from price. b. Segments: Data Center $4.8 bn (+61% QoQ, outpacing the market). Edge $5.8 bn (+7% QoQ, resilience in premium and mainstream, contraction in low-end). Consumer $397 mn (-29% QoQ, channel still digesting price hikes). c. Margins: Non-GAAP GPM 85.1% (vs. 84.6% in the prior quarter, guide 83%–85%), driven by both volume and price. OPM 80.4% (vs. 79.2% in the prior quarter).

d. Opex and EPS: Opex $513 mn, or 4.7% of revenue. EPS $49.51, vs. guide of $44–46. e. GAAP adjustments: SBC $91 mn; gains on equity securities $513 mn, mainly from holdings in Nanya Tech.

4. Cash flow and capex

a. Operating cash flow $6.4 bn; Adj. FCF $5.7 bn, with ~52% margin. Period-end cash $5.1 bn, with no debt. b. Total capex $378 mn, or 3.4% of revenue. It will step up for the rest of the fiscal year, with FY27 at a mid-single-digit percentage of revenue.

II. Details from the call

2.1 Management highlights

1. NBM (long-term new biz. model contracts)

a. Two NBM contracts finalized this quarter were already disclosed last call. The company is negotiating extensions with existing customers and new contracts with new customers. b. The optimal NBM bit share is slightly above the current level, and the company will be patient in selecting long-term partners. The NBM pipeline exceeds the company's supply through the end of this decade, and customer signals extend beyond FY2030.

c. Recent engagement with leading global AI companies indicates further upside in demand. 2. Technology roadmap

a. BiCS 8 is the current workhorse node, introducing the CBA (CMOS-to-Array direct bonding) architecture. It meets the performance needs of AI data centers. b. BiCS 9 reuses the BiCS 8 storage array with a next-gen CMOS, materially improving interface performance. c. BiCS 10 is a full-node upgrade: 1Tb TLC dies target performance workloads, and 2Tb QLC dies lead in density, with read/write bandwidth at 2x that of BiCS 8.

d. HBF: The first HBF storage die completed tape-out this quarter and moved into manufacturing. The company still plans to sample initial products next calendar year; HBF is not included in the Investor Day long-term financial model and will be discussed once customer onboarding and manufacturing readiness are clearer.

3. AI demand and end markets

a. KV cache in inference keeps expanding, with flash serving the capacity tier and avoiding costly recomputation. Lab tests show SSD-based systems deliver ~3x token throughput vs. DRAM-only designs, with ~75% lower energy use. b. By the end of this decade, KV cache could account for ~1/3 of AI data center flash. The AI data center opportunity could approach the size of today's entire NAND industry.

c. Data centers' bit share in the industry has risen from ~20% early this decade to about half. This quarter, the company allocated more supply to data centers, including three US hyperscalers. d. Edge: Auto and IoT customers are growing strongly, and auto could become one of the largest markets within Edge. The consumer brand connects directly with hundreds of millions of end users, providing early demand signals.

e. Industry outlook: CY2027 is set to be strong, with data centers at 55%+ of the market and bit demand up ~40% YoY (CY2026 was up by more than 2x).

2.2 Q&A

Q: Will NBM signings continue at scale? Will the bit share covered by contracts change over the next few years?

A: Approx. 50% of FY27 bits and about two-thirds of FY28 bits are under NBM, and the ratio will tick up. The pipeline is robust and still expanding, and more NBMs are expected over the next several quarters, as NBM has become the primary collaboration model with customers. About 95% of counterparties are investment-grade, with a combined operating cash flow of roughly $130 bn.

Multiple customers said they underestimated NAND needs and came back for a second round, which must be balanced against onboarding new customers. Among the largest customers, one has already signed a second tranche. Q: PC vendors are pressured by memory price hikes. Will there be NBM demand in PCs?

A: No customer-specific comments; NBM is centered on data centers, with some Edge exposure, and is open to all customers. The company expects PC unit shipments to decline this year and to edge down slightly next year (low single-digit), and the current two calendar quarters mark the worst YoY trough.

That said, EB content per PC will still rise next year. Internal models show PCs, smartphones, and data centers all growing next year, driven by higher per-box capacity and, in part, mix upgrade. Q: Data center revenue was up nearly 20x YoY and +61% QoQ. What drove the penetration gains?

A: The BiCS 8 node, the right product set, and more granular NAND usage in inference architectures. CBA delivers better interface speeds; TLC products have been in market for years, while QLC only started contributing a few quarters ago and is ramping faster than expected.

Over the past quarter, customers optimized TLC vs. QLC across KV cache, staging, high-speed data lakes, and pre-/post-training, with some QLC now entering KV cache. SLC is entering architectural discussions (no sales yet), valued for endurance as a DRAM substitute, with a large price gap and strong stickiness once adopted. Q: Over the long run, can the company's data center bit mix catch up with the industry's ~half?

A: We are working to match the industry mix and will push through CY2027. Timing of a full catch-up remains to be seen, but the share will be meaningfully higher vs. FY26. Q: How do volume and price contribute to next quarter's guide? What is the price trajectory?

A: No split by volume vs. price, but both contribute next quarter. Prices will continue to rise modestly throughout the year. Q: Any updates on HBF engagement with customers and partners?

A: Active discussions with potential customers, with no new quantified disclosures. The next step is to proceed with sampling so customers can place HBF in their systems. Q: Will new NAND layers such as Storage Next and CMX near-memory all win?

A: We think multiple approaches will be tried, and customers deploying capital will decide by use case. As inference scales, architectures will keep optimizing for economics, pulling NAND into more stages. Historically, the direction was TLC to QLC for density, but there is now reverse demand as well: using NAND's endurance and performance in certain stages to replace pricier and scarcer tech, and we are deeply engaged in customer architecture discussions. Q: Will you accelerate BiCS 8→9→10 migration to boost near-term bit growth?

A: No. We maintain a mid-to-high-teens bit growth plan. Node migration requires planning, tool buys, and line scheduling; fab plans are set to that growth pace, and there is no change at this time. Q: Is there a cap to tilting supply toward data centers?

A: There are constraints; in the short run, adjustments are limited, while over time flexibility rises. The TLC/QLC output mix per wafer has set ratios, and we must continuously match inventory, output, and customers; a wafer cycle is ~120 days. Q: With the NBM pipeline exceeding supply through decade-end, does this change your supply view or capex plans (the questioner estimates mid-single-digit capex implies ~$2.5–3.0 bn)?

A: The pipeline extends to 2030 and beyond, but it is too early to adjust supply based on it. The model launched only two to three quarters ago, and post-Investor Day customers proactively proposed NBM; NBM is not the only biz. model. We may adjust supply-demand later, but that requires negotiations and firm commitments, and we did not address capex changes.

Q: Can NBM collateral be used for buybacks, or is it restricted cash?

A: In theory yes, with no legal constraints, but most of it is off-balance-sheet financial guarantees. The guarantees are held by third-party financial institutions; on the balance sheet, only prepayments exist and they are minimal. Q: With agentic AI adoption, what is the 2027 enterprise SSD capacity per server?

A: No server-level figure; for agent orchestration, it is ~60–90GB per CPU. This demand ties to CPU count and has not been mapped to server TAM and elasticity; follow-up will be provided. NAND is entering inference stages via TLC, QLC, and high-endurance options, and some of it can substitute DRAM. Q: Can enterprise SSD share reach the low-to-mid teens by 2027?

A: No share target, but data center mix will keep rising in 2027. Share targets can skew portfolio choices; we must balance short-, mid-, and long-term value. Data center was up ~17x YoY, and the balance point depends on quarterly dynamics and subsequent NBM signings. Q: Will BiCS 8/9/10 migration help meet data center bit needs?

A: BiCS 9 is key to the data center strategy, and BiCS 10 is still further out. With CBA, BiCS 9 boosts interface speed on the same cell array and will have a measured ramp; QLC is scaling, and BiCS 8 continues to ramp. Q: Peers mentioned SLC NAND. How will it evolve next year, and will it tighten NAND supply?

A: Too early for concrete forecasts, but worth close attention. Some customers have signaled SLC demand; making SLC vs. TLC/QLC has about a 3:1 substitution ratio and will affect overall supply, with magnitude depending on actual deployment. It is still at the architecture-discussion stage; updates will come in the next few quarters. Q: For the Mar quarter (FY27 Q3), do you expect QoQ revenue up, down, or flat?

A: No Q3 guide; model QoQ growth from Q2 to Q3 conservatively. Q3 is historically the weakest sequential quarter; for example, bits once fell 19% QoQ. Q: After the buyback authorization is used up, will the BOD renew each quarter? How will you execute if the stock pulls back?

A: We remain committed to returning 100% of excess cash to shareholders, and the BOD meets quarterly; authorization is not a constraint. We have repurchased about 4% of float over the past two quarters and will continue.

Q: For cloud customers, what is the TLC vs. QLC mix, and will it change materially later this decade under NBM?

A: The mix is currently ~70:30 TLC:QLC and should broadly hold through this fiscal year. There will be minor quarterly swings; customers indicate QLC is re-accelerating across use cases, and future shifts depend on ongoing discussions. The internal 'Stargate' QLC products have completed all qualifications and are ramping at customers.

Q: Handset market declines by double digits this year. Can your handset bits grow next year?

A: No forecasts for a single market; we only note EB per handset should rise next year. The low end is under pressure, causing a slight decline in units, but capacity per device is rising; we will manage mix dynamically to optimize near-, mid-, and long-term financial outcomes. Q: Do 'top AI companies' mean AI labs or new-age cloud providers? Is there credit risk contagion from the latter?

A: No credit issues observed; demand strengthened vs. last quarter. The largest customers are mostly cash-generative investment-grade names. When pressed on types, we only said customers span hyperscalers and model providers, among others, without naming specific companies.

Q: AR up — is it due to more bits under NBM changing payment terms?

A: No; it is primarily a customer-mix effect. There was a minor impact from the intra-quarter revenue cadence. Q: With a more predictable model, will you add leverage to accelerate capital returns?

A: No change near term: stay debt-free and return 100% of cash flow to shareholders after R&D. As laid out at Investor Day, the financial model runs through 2030; we will continue to invest in HBF, 3D Matrix storage, and the BiCS roadmap. Q: What are the variables for HBF capacity growth? What milestones lead from sampling to scale revenue?

A: It is too early to size the TAM; the key milestone is putting physical product in customers' hands. Next steps include taping out the base die (controller), assembling with storage dies, and enabling customer integration and validation within their infrastructure instead of simulation. Customer feedback will inform market size, ramp speed, and architectural placement.

We are deliberately limiting touchpoints to avoid overloading engineering teams. Once customers have hardware, we will broaden use-case discussions. Q: Will SLC needs from more granular inference architectures cannibalize TLC/QLC?

A: We do not view it as cannibalization; rather, TAM is expanding into new attributes. Wafers can be run in SLC mode with high flexibility; it is still early. The key is whether SLC endurance and performance can replace some DRAM, where the cost gap is ~20–40x; NAND is the most scalable semiconductor technology, and scalable tech attracts TAM. Q: While awaiting HBF commercialization, can the JV fab add SLC capacity to meet near-term demand?

A: There is no standalone SLC capacity; existing wafers can run in any mode. Cells can switch among SLC, MLC, and TLC, so we can ship using existing wafers; it is too early to discuss supply. Customers are still exploring use cases and demand levels, which will influence the market and warrant attention.

<End of text>

Risk disclosure and statements for this article:Dolphin Research Disclaimer and General Disclosure