SanDisk at Citi’s 2026 Global TMT Conference: betting on ai demand

SanDisk said data-center revenue reached $3.0B, or 38% of quarterly sales, generated $5.0B of free cash flow, and returned $4.5B to shareholders through buybacks. Management targets 50% of NAND bits under long-term NBM contracts by fiscal 2028, with contract floor gross margins around 80%, and expects mid-to-high-teens revenue and bit growth through fiscal 2030. The company sees AI-driven NAND TAM expanding from a historical $50B to as much as $500B by 2027, while HBF inference-memory products are expected to reach customers in 2027; key risks remain supply discipline, execution and competition from YMTC.

Analysis

The investable issue is not near-term NAND demand but whether SNDK can structurally reduce its historic earnings beta to spot pricing. Contracted volume with pricing bands can raise the durability of trough cash flow and justify a higher multiple, but it also transfers a meaningful portion of a NAND upcycle’s upside to customers. The market should value SNDK less as a pure commodity-memory torque vehicle and more as a hybrid: contracted data-center supply plus residual spot optionality. Evidence to underwrite that rerating is disclosed contract duration, take-or-pay/financial guarantees, and the share of gross profit—not merely bits—covered by NBMs.

The most important second-order risk is that industry supply discipline is not controlled by SNDK. Kioxia JV capital decisions, Samsung/SK Hynix capacity actions, and YMTC’s ability to move from components into enterprise SSDs determine whether current margins persist. A higher data-center mix also raises customer concentration and negotiating leverage: hyperscalers signing long-term supply agreements may secure volume precisely when SNDK has the greatest pricing power. Over the next 1-3 months, contract disclosures and peer capex commentary should matter more than conference optimism; over 6-18 months, HBF remains a zero-revenue technology option until qualification converts into customer purchase commitments.

Contrarian view: the headline cash return may be less bullish than it appears if it reflects a cyclical peak in NAND pricing rather than sustainably lower capital intensity. Buybacks are accretive only if normalized FCF remains high after JV investment and node-transition spending. Conversely, consensus may underappreciate the strategic value of becoming a preferred supplier to AI infrastructure buyers: qualification cycles and supply assurance can create stickiness that a spot-market model never offered.

SNDK is a watch-list long rather than a chase given macro rate pressure and the lack of independently verifiable NBM economics. The clean catalyst is the next earnings release demonstrating stable or rising gross margin while contract coverage expands and capex/revenue remains contained; a material cut to bit-growth, rising capex intensity, or margin pressure despite favorable pricing would invalidate the rerating thesis.

Key Decisions for Investors

• Initiate a starter long SNDK only on post-earnings confirmation that contracted business is expanding without a reduction in spot-price participation; scale over 1-3 months. Target a multiple rerating if management discloses durable contracted gross-profit coverage, not simply contracted bit volume. Risk: NAND spot pricing rolls over or capex rises faster than revenue.
• Express the durability thesis as long SNDK / short WDC over 6-12 months, sized modestly. SNDK’s potential contract-driven cash-flow visibility and data-center mix improvement should outperform a more cycle-sensitive legacy storage exposure; close if SNDK’s margin premium narrows for two consecutive quarters or WDC demonstrates comparable long-term pricing protection.
• Maintain a sector supply-discipline dashboard: alert on Kioxia JV capex expansion, Samsung/SK Hynix NAND capacity additions, or evidence that YMTC is winning enterprise-SSD qualifications. Any broad increase in planned wafer starts is a signal to reduce SNDK exposure before the 6-18 month pricing impact reaches reported results.
• Do not assign material value to HBF before 2027 customer qualification data. Treat announced ecosystem participation by GOOG, META and SKHY as validation of technical relevance, not a demand commitment; upgrade the thesis only upon disclosed design wins, volume commitments, and incremental-capex requirements.