Sandisk: A Sandcastle by the Beach; Initiate NEUTRAL

Summary

We initiate Sandisk with a NEUTRAL rating and a target price of USD1,582, based on 7x FY2029E EPS of USD225.94, implying 15.8% upside. NAND prices are the biggest threat to Sandisk. Its FY2026 revenue growth came mostly from higher prices rather than volume, and about 75% of sales still come from price-sensitive Edge and Consumer products. With the market expecting supply to overtake demand in 2027, a 30% fall in blended prices would cut gross profit by about 35%, even if gross margin stayed near 78%. AI inference demand, multiyear customer agreements and shared manufacturing with Kioxia improve the business, but none protects Sandisk from a price downturn. We need evidence that earnings can hold up as NAND prices fall before assigning a higher multiple.

Business overview

Separated from Western Digital in February 2025, Sandisk designs NAND flash memory, SSD controllers and finished storage products. Datacenter supplies enterprise SSDs for cloud, conventional servers and AI workloads. Edge serves smartphone, PC, gaming, automotive and industrial device makers, while Consumer comprises retail memory cards, USB drives and portable SSDs. Sandisk competes with Samsung, Micron, SK hynix/Solidigm and Kioxia, and has a strong retail brand but a smaller enterprise SSD franchise.

Sandisk's share price has surged 838% over the past 1 year as one of the major beneficiaries of the AI-driven NAND upcycle. Surging demand for AI infrastructure and enterprise storage, coupled with constrained industry supply, has pushed NAND selling prices sharply higher across end markets. This combination of stronger shipments and exceptional pricing lifted FY2026 revenue 175% to USD20.25bn. Datacenter sales grew 437% to USD5.15bn on almost 120% exabyte shipment growth, showing genuine physical demand. By contrast, Edge revenue rose 195% on only high-single-digit exabyte growth, while Consumer revenue rose 29% despite a mid-teens volume decline. Datacenter has become the clearest shipment growth engine, but Edge and Consumer still generated roughly 75% of FY2026 sales, leaving consolidated earnings highly exposed to NAND selling prices.

Figure 1: Sandisk’s 1y performance

Figure 2: Revenue rose fivefold in a year as gross margin tripled

Table 1: Revenue growth greatly exceeded physical shipment growth

Segment	FY2025 revenue (USD mn)	FY2026 revenue (USD mn)	Revenue YoY	Exabyte shipments YoY	Revenue per GB YoY
Datacenter	960	5,153	437%	Roughly +120%	Roughly +150%
Edge	4,127	12,160	195%	High single digits	Roughly +180%
Consumer	2,268	2,935	29%	Mid-teens decline	Low fifties % increase
Source: Sandisk 10-K, iFAST compilations. Data as of 21 October 2026.

Sandisk makes two main types of NAND. TLC stores three bits per cell and balances speed and endurance, suiting phones, PCs and performance SSDs. QLC stores four bits per cell, lowering cost per terabyte but wearing out faster and slowing under sustained writes. QLC therefore suits read-heavy, high-capacity storage such as AI data lakes, where controller and firmware design decide whether drives pass enterprise qualification.

 Figure 3: Each extra bit per cell cuts endurance

Shared fab economics, not an asset-light business

Substantially all of Sandisk's NAND wafers come from Flash Ventures, its manufacturing joint ventures with Kioxia in Japan. Sandisk owns 49.9% and Kioxia 50.1%. The partners jointly develop BiCS NAND and each generally receives about half of wafer output. Kioxia owns and operates the fab buildings, while the joint ventures own or lease the production tools. Each partner pays variable costs for its share of wafers, but Sandisk bears about half of the joint ventures' fixed costs regardless of how many wafers it buys. It also funds 49.9% to 50% of equipment investment when joint-venture cash flow falls short, and guarantees part of the equipment lease obligations.

Sandisk reported only USD177mn of direct capex in FY2026, but this understates the cost of its manufacturing. Flash Ventures is equity-accounted, so the joint ventures' equipment spending sits outside reported capex. Sandisk funds its share through wafer purchases, loans, capital contributions and lease guarantees instead. The Kioxia partnership shares the investment burden but does not make Sandisk asset-light, and the fixed-cost obligation limits how far it can cut costs in a NAND downturn.

Industry overview: A larger NAND market can still be cyclical

AI servers increasingly use enterprise SSDs to stage data, store checkpoints, retrieve documents and retain inference context. Flash fills a capacity and cost tier between expensive DRAM and slower HDDs, rather than replacing HBM. Sandisk estimates data centres will take about half of industry bits in 2026, but consumer devices remain material and price sensitive. Industry NAND bit demand grew every year from 2015 to 2024, yet blended ASP fell in eight of those ten years. More NAND consumption has not historically guaranteed manufacturer pricing power.

Today’s exceptionally high prices are the source of the risk. Forecasts show industry ASP rising 267% in 2026 and a further 10% in 2027, but these are annual averages. Because prices rose significantly in 2026, the 2027 average can still rise even if prices fall every quarter during the year. We therefore watch quarterly spot and contract prices, which will show a downturn first. The question is not whether NAND is in shortage today, but how far prices fall once it ends.

Figure 4: NAND bit growth remained positive despite volatile ASP growth

Supply can respond before new fabs are built. Higher layer counts, a greater QLC mix and better yields raise bits per wafer even when wafer starts barely change. Sandisk targets about 27% annual improvement in bits per wafer, faster than its planned mid-to-high-teens bit shipment growth. Bloomberg Intelligence similarly projects strong NAND bit-production growth before wafer capacity expands more sharply in 2028 and 2029. This productivity helps each supplier but undermines industry pricing when every competitor achieves it. YMTC and other Chinese suppliers add further capacity, first in commodity products but eventually across the global NAND market.

Figure 5: Global NAND bit production is expected to slow

Figure 6: Wafer-start capacity expansion by NAND manufacturer

The market is estimating that the shortage will ease in 2027 but differs on how far. TrendForce’s July 2026 outlook shows NAND moving from undersupply in 2026 to surplus in 2027, while DRAM stays short. Gartner’s September 2026 outlook instead expects the shortage to narrow from about 9% in 2026 to 5% in 2027. The direction is the same in both, and Sandisk’s earnings will fall with prices once the shortage ends as the pricing power shrinks. Only the timing is uncertain.

Supply can also rise faster than fab announcements suggest, because what matters is bits, not wafers. Switching from TLC to QLC or improving yields lets a supplier sell more storage from the same fab, while new capacity takes time to equip and qualify. That is why today’s shortage can sit alongside a growing risk of oversupply, and why we focus on bits actually shipped rather than capex announced.

Figure 7: TrendForce projects NAND supply moving from deficit to surplus in 2027

Source: TrendForce

The balance could also diverge across products. Demand for high-capacity enterprise SSDs may stay strong as AI infrastructure expands, while smartphone and notebook demand is more price-sensitive. If supply catches up, price pressure may appear first in consumer and client products rather than qualified data-centre SSDs. With Edge and Consumer generating most of Sandisk's FY2026 revenue, growth in AI exabytes alone would not shield group earnings from a broader price correction.

AI inference provides demand optionality

AI inference is creating new opportunities for NAND as models process longer conversations, retrieve larger datasets and require more memory to store model parameters and contextual information. One industry scenario expects agentic AI's share of global NAND shipments to rise from 2% in 2026 to 22% by 2030. However, improvements in KV-cache compression and memory management could reduce NAND consumption per inference task, while HDDs remain more economical for less frequently accessed data.

Figure 8: Agentic AI's projected share of global NAND shipments increases from 2% in 2026 to 22% by 2030

Looking further ahead, Sandisk is developing High Bandwidth Flash (HBF) to address the growing memory-capacity constraints of AI processors. Currently, GPUs rely on HBM for high-speed computation, but HBM's high cost and limited capacity make storing large AI models expensive. HBF aims to introduce a much higher-capacity NAND-based memory tier closer to the GPU, allowing large model weights and other read-intensive data to be stored near the processor while HBM continues handling latency-sensitive workloads. Sandisk's proposed architecture illustrates GPU memory capacity increasing from 192GB using HBM alone to over 3TB through a combination of HBM and HBF, potentially reducing the cost of serving larger AI models.

However, these configurations remain conceptual rather than commercially deployed systems. Sandisk targets its first HBF inference product samples in 2027, with adoption dependent on achieving the required bandwidth, latency, endurance and manufacturing economics. We therefore view HBF as a potential expansion of NAND's addressable market rather than an established competitive advantage and exclude meaningful HBF revenue from our FY2029 forecasts.

Figure 9: Conceptual expansion of the AI memory stack

Source: MSUSA estimates.

Figure 10: HBF is a credible technological opportunity but has yet to translate into commercial revenue

Source: Sandisk Investor Presentation

Thesis 1: NAND prices are the biggest threat to earnings

In calendar 1Q26, Sandisk’s blended NAND ASP rose about 140% QoQ while its bit shipments fell, the most price-driven growth among the major suppliers. Peers also raised prices sharply, often with stronger bit growth, which points to an industry shortage rather than Sandisk-specific pricing power. Differing fiscal quarters, mix and contract timing mean one quarter cannot prove a lasting disadvantage, but the dependence on price is clear.

Figure 11: Sequential NAND bit shipments versus ASP change across leading suppliers

FY2026 Datacenter exabyte growth was genuinely strong, but Edge and Consumer grew almost entirely on price. Sandisk's 4Q26 gross margin reached 84.6%, against 26.4% a year earlier and 7.1% in FY2023,highlighting how much profitability has benefited from exceptionally tight NAND supply. As illustrated in Figure 7, the projected narrowing of the NAND supply-demand gap could weaken pricing power over the next two to three quarters, even without outright oversupply. While manufacturing cost per bit should also decline as technology improves, margins would come under pressure if ASPs fall faster than costs. Unlike Micron, SK hynix and Samsung, Sandisk has no DRAM or HBM profit pool to cushion the blow.

The right operating test is how much revenue growth comes from exabyte shipments and enterprise mix once ASP is stripped out. A rising Datacenter share is not enough if the revenue comes mainly from unusually strong prices. Conversely, Sandisk could outperform even in a modest correction if it wins more qualified enterprise capacity and lowers its cost per bit.

Thesis 2: Product competitiveness has not yet established a defensive moat

Sandisk’s enterprise SSD footprint is expanding from a relatively small base. In calendar 2Q26, it generated approximately USD2.98bn of enterprise SSD revenue versus USD14.35bn for Samsung and USD8.63bn for SK hynix/Solidigm. Its share of the five-vendor revenue total was roughly 8%, which is not equivalent to its overall NAND bit-manufacturing share. The gap provides a growth runway but also reflects entrenched competitors with broader enterprise qualifications and established hyperscaler relationships.

Figure 12: Sandisk’s enterprise SSD revenue scale remains below leading suppliers

Sandisk's UltraQLC platform writes directly to QLC, reducing the slowdown when a drive's pseudo-SLC cache fills. BiCS9 pairs a BiCS8 array with newer CMOS logic, and BiCS10 targets about 60% higher bit density. These are commercially relevant, but Micron, Solidigm, Samsung and Kioxia offer similar high-capacity QLC products and controller technology. Kioxia is particularly instructive, as the two companies share BiCS development and manufacturing while competing for the same SSD customers. Sandisk therefore cannot claim exclusive wafer technology or a proven, industry-leading SSD cost structure. Hyperscalers typically qualify several vendors on endurance, latency, firmware, reliability and total cost of ownership. Incremental enterprise orders are achievable, but lasting price premiums remain unproven.

Table 2: Competing NAND and enterprise SSD product roadmaps

Supplier	Technology / enterprise focus	Competitive implication
Sandisk	BiCS CBA NAND; UltraQLC, 128TB / demonstrated 256TB	Strong capacity roadmap, but large-scale 256TB deployment remains to be proven
Micron	G9 QLC; 6600 ION 245TB commercially shipping	Comparable ultra-high-capacity product already in market
SK hynix	321-layer NAND, established enterprise TLC and QLC portfolio	Mature QLC enterprise qualifications and installed base
Samsung	V-NAND; PCIe 5.0 enterprise TLC / QLC portfolio	Large manufacturing scale and broad enterprise product range
Kioxia	Joint BiCS technology; 245TB-class QLC qualification	Shared wafer technology with Sandisk, competes downstream

Since Sandisk's rapid growth was benefited from AI-driven demand exceeding available supply rather than a proven competitive advantage. As supply constraints ease, its smaller enterprise SSD footprint and limited pricing differentiation could weaken earnings resilience, limiting the case for a higher valuation multiple.

Thesis 3: Long-term agreements and Kioxia’s collaboration reduce risk but do not guarantee peak margins

Sandisk's New Business Model (NBM) agreements improve on spot-linked purchasing. Ten agreements with eight customers cover about half of FY2027 NAND bits and two-thirds of FY2028, with an average term of more than four years. The USD93.9bn minimum contractual revenue is backed by binding purchase commitments and financial guarantees, while pricing floors support management's expectation of approximately 80% gross margins on contracted products. These agreements provide meaningful earnings protection and represent a structural improvement over previous NAND cycles.

Table 3: Sandisk’s New Business Model agreements

Metric	3Q26 update	4Q26 update
NBM agreements announced / signed	5 total: 3 signed in 3Q26, 2 signed so far in 4Q26	10 total agreements with 8 customers
Contract duration	Longest up to 5 years	Up to 5 years; weighted average over 4 years
FY2027 bit coverage	More than one-third of bits	More than 50% of bits
FY2028 bit coverage	Not disclosed	Approximately two-thirds of bits
Minimum contractual revenue	Approximately USD42bn from three 3Q26 contracts	USD93.9bn minimum across all signed NBMs
Remaining performance obligations (RPO)	Not disclosed	USD59.8bn at quarter-end; USD91.1bn including 2 post-quarter NBMs
Financial guarantees	More than USD11bn across 5 agreements	USD16.5bn across all NBMs
Source: Company filings, Bloomberg Intelligence.

However, contracted bit coverage does not represent the share of group revenue or gross profit protected at today's prices. Only 19% of the USD59.8bn year-end remaining performance obligations (RPO) is expected to be recognised within 12 months, excluding two post-quarter agreements. More importantly, nearly half of FY2027 bits and one-third of FY2028 remain outside the NBMs. As discussed in Thesis 1, weaker NAND pricing could pressure earnings from these uncontracted volumes, even if customers fulfil their commitments. Management's approximately 80% margin expectation also applies to contracted products rather than the entire business.

Manufacturing adds a further constraint. Sandisk bears roughly half of Flash Ventures' fixed costs regardless of wafer offtake and shares in joint capital investment. Underutilisation charges of USD249mn in FY2024, USD75mn in FY2025 and USD11mn in FY2026 show that idle capacity has a cost. Joint-venture coordination, binding wafer orders and concentration in Japan all reduce flexibility in a downturn. This does not make Kioxia a bad partner, but shared capex is not the same as a flexible, asset-light cost base.

Reported FY2026 free cash flow of about USD11.5bn also exceeded adjusted FCF of about USD8.7bn, which strips out customer advances, deposits and Flash Ventures flows. While underlying cash generation remains strong, future FCF sustainability will depend increasingly on cash generated from product deliveries rather than upfront customer payments.

Overall, we believe NBMs represent a structural improvement in Sandisk's earnings resilience. However, uncontracted NAND exposure and uncertainty over profitability beyond existing contract terms remain important considerations. We therefore see insufficient evidence to capitalise today's exceptional group-wide margins into a permanently higher valuation multiple.

Valuation: Initiate Neutral, with less favorable risk to reward ratio

Management's long-term framework targets mid-to-high-teens annual revenue growth, a non-GAAP gross margin of about 80% and an operating margin of 75% over FY2028 to FY2030, supported by contracted demand and better manufacturing efficiency. This implies profitability far above previous NAND cycles despite a likely supply recovery. Multiyear agreements improve visibility, but the ability to sustain such margins is untested. We see little case for a higher multiple until Sandisk shows its earnings can hold up as NAND prices normalise.

Table 4: Management guidance

Metric	FY2026 actual	4Q26 actual	1Q27 guidance	FY2028–FY2030 target
Revenue (USD bn)	20.25	8.97	10.30–10.80	—
Revenue growth	+175% YoY	+372% YoY	+14.9–20.5% QoQ	Mid-to-high teens annually
Gross margin (non-GAAP)	71.60%	84.60%	83–85%	80%
Operating expenses (non-GAAP)	USD1,782mn	USD484mn	USD520–540mn	5% of revenue
Diluted EPS (non-GAAP)	USD70.88	USD39.25	USD44–46	—
Adjusted FCF margin	43.20%	56%	—	~50%
Source: Sandisk 4Q26 financial results (5 August 2026), Sandisk Investor Day (13 August 2026), iFAST compilations.

We value Sandisk at 7x FY2029E EPS of USD225.94, giving a 12-month target price of USD1,582, 15.8% above the latest close of USD1,365.38. We set the multiple at the middle of the 6x to 8x range for memory makers and below our 10.5x for Micron, reflecting Sandisk's greater dependence on NAND pricing and lack of a DRAM or HBM earnings buffer.

Table 5: Valuation summary

	FY25A	FY26A	FY27E	FY28E	FY29E
Revenue (USD mn)	7,355	20,248	48,936	58,318	51,248
Revenue growth YoY		175.3%	141.7%	19.2%	-12.1%
EPS (USD)	-11.32	73.76	212.62	260.50	225.94
Earnings growth YoY			188.3%	22.5%	-13.3%
P/E	27.8x	24.7x	6.4x	5.2x	6.0x
Current price (USD)					1,365.38
Fair P/E on FY29E (input)					7.0x
Target price (USD)					1,582
Upside potential					15.8%
Source: Bloomberg Finance L.P., company filings, iFAST Compilation. Data as of 21 October 2026.

Risk

1. NAND supply recovery and pricing pressure. Faster supply growth from technology upgrades, new capacity and Chinese makers could push NAND into oversupply from 2027. Because Sandisk's recent growth came mostly from price, falling ASPs would hit its earnings harder than those of diversified peers.

2. Limited protection from long-term contracts. Although multiyear agreements improve revenue visibility, contractual price floors may not fully protect earnings during a NAND downturn. Contract renegotiations or weaker realised ASPs could further pressure margins.

3. Enterprise SSD competition and execution. Strong competition from Samsung, Micron, SK hynix/Solidigm and Kioxia could limit Sandisk's market-share gains. Slower QLC adoption or customer qualification could constrain shipment growth.

4. AI adoption and emerging competition. Slower hyperscaler spending, improved memory efficiency and HDD cost advantages could constrain NAND demand. Meanwhile, YMTC's expanding supply and uncertainty surrounding HBF commercialisation pose additional risks.

Key takeaway

Sandisk is benefiting from real AI storage demand, with strong Datacenter growth and multiyear customer agreements providing greater earnings visibility than previous NAND cycles. However, we believe the benefits of these contracts have been largely priced in following the stock's substantial re-rating, while much of its exceptional profitability remains linked to elevated NAND selling prices. Further upside will depend on Sandisk's ability to sustain earnings beyond the current upcycle. Given the less favourable risk-reward profile, we initiate coverage with a NEUTRAL call and a target price of USD1,582.

Disclaimers:
This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Capital does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.
For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.