UBS Initiates Coverage on Kioxia: AI Inference Drives Long-Term NAND Upcycle; Maintain Buy

Overview

UBS publishes its first coverage report on Kioxia Holdings (Kioxia Holdings, 285A.T), assigning a Buy rating and a target price of ¥79,000. The central thesis is that the market has not yet fully priced in the structural shift in NAND demand driven by AI inference or Kioxia’s distinctive cost advantages. As AI applications shift from training to inference, NAND is evolving beyond simple data storage into a critical computing component for managing KV Cache. This results in highly inelastic demand. Meanwhile, competitors’ capital expenditures have been redirected toward high-margin HBM and advanced DRAM, constraining NAND capacity expansion. UBS expects supply-demand tightness to persist through 2027, with NAND prices rising sequentially over the next six quarters. Leveraging its leadership in lateral scaling technology and joint venture scale, Kioxia maintains the industry’s lowest bit cost—positioning it to sustain superior margins throughout an extended upcycle.

Core views

Demand Side: AI Inference Reshapes NAND’s Role. Traditionally used for PC and mobile storage, NAND is now integral to AI inference systems—specifically for retaining user conversation history (KV Cache). As context windows expand and RAG (Retrieval-Augmented Generation) and Agent Search gain traction, KV Cache grows exponentially—far exceeding HBM capacity limits—necessitating offloading to high-performance NAND SSDs. This architectural shift ties NAND demand linearly to GPU deployment volumes and renders it price-insensitive. Even with compression techniques (e.g., TurboQuant), offload demand remains rigid, as context-length growth outpaces HBM capacity doubling rates. Supply Side: Capacity Expansion Lagging; Shortage Extends Through 2027. Major memory vendors (Samsung, SK hynix, Micron) have shifted capex and R&D focus toward high-profit HBM and advanced DRAM, causing NAND wafer capacity to peak in 2022 and decline thereafter. Although bit density improves via technology advances, aggregate bit supply growth lags behind demand growth. UBS forecasts sequential NAND price increases over the next six quarters, peaking in Q3 FY2027. Company Competitiveness: Kioxia’s Cost Moat. Since 2022, Kioxia has maintained the industry’s highest operating margin—driven by its unique lateral scaling technology path, rather than solely relying on vertical stacking layers. Combined with CBA (CMOS Bonded Array) technology, Kioxia achieves wafer manufacturing costs of ~$2,000—significantly below Samsung (~$4,000) and Micron/SK hynix (~$6,000). Additionally, its joint venture with SanDisk enables shared facilities and R&D, delivering economies of scale and high yields—allowing Kioxia to match Samsung’s bit shipment share despite lower capex. Valuation & Outlook: UBS derives the ¥79,000 target price using FY2028 forecast book value per share (BPS) and a 4.7x PBR multiple, implying a 4.0x EV/EBITDA multiple. This valuation reflects NAND’s transition—from a volatile, low-free-cash-flow cyclical industry to a structurally growing, cash-flow-stable sector powered by AI. UBS projects Kioxia’s operating profit at ¥7.11 trillion and ¥8.99 trillion for FY2027 and FY2028 respectively—substantially above consensus.

Analysis framework

UBS’s analytical logic follows a sequence: macro-industry trends → micro-competitive landscape → company fundamentals → valuation re-rating. First, by deconstructing AI inference architecture (KV cache offloading, multi-tiered storage systems), the report demonstrates a fundamental shift in NAND’s demand nature—from peripheral storage to core computing component—thereby challenging conventional ‘silicon cycle’ concerns. Second, on the supply side, it contrasts peers’ capex allocation (HBM vs. NAND) to infer structural bottlenecks in NAND capacity expansion. Third, diving into technical detail, it highlights how markets often misjudge competitiveness based on stacking layers alone, overlooking lateral scaling and wafer cost as true determinants of profitability—thus revealing Kioxia’s underappreciated cost advantage. Finally, for valuation, it applies a PBR framework anchored in ROE and cost of equity, while referencing historical EV/EBITDA ranges and peer comparisons to justify that current pricing fails to reflect the industry’s structural improvement in earnings quality.

Methodology notes

• Industry/sector analysis frameworkSupply-demand framework

  Supply-demand balance and price cycles

  The report assesses price trends by analyzing the gap between NAND bit supply growth (constrained by capacity and technology) and AI-driven demand growth. It notes that when supply expansion is hampered by capex priorities (e.g., shifting to HBM), prices will continue rising due to shortages—even amid strong demand—until a new equilibrium emerges.

• Competition and strategy frameworkMoat / competitive advantage

  Cost leadership strategy (bit-cost advantage)

  The report emphasizes Kioxia’s structural cost advantage—achieved through lateral scaling technology and JV scale effects—delivering the industry’s lowest wafer and bit costs. This cost moat enhances profit elasticity during price-up cycles and secures competitive leadership.

• Valuation methodologyPB valuation

  PBR = ROE / CoE

  The report anchors valuation in price-to-book (PBR), deriving a fair multiple via PBR = Return on Equity (ROE) / Cost of Equity (CoE). Using a long-term average ROE (41%) and CoE (8.6%), it calculates a 4.7x PBR to smooth short-term volatility and reflect sustainable profitability.

• Industry/sector analysis frameworkUpstream-midstream-downstream value chain transmission

  AI architecture transformation’s impact on upstream memory

  The report analyzes how AI inference architectures (e.g., RAG, Agent Search) drive exponential KV Cache growth, forcing data offload from HBM to NAND SSDs. This downstream application shift directly transmits upstream—altering NAND’s product attributes (from storage to compute-assist) and demand drivers.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Kioxia Holdings (285A.T)

  Primary beneficiary. As a pure-play NAND vendor, it benefits directly from AI inference-driven NAND demand surges and price increases—and possesses the industry’s lowest cost structure.

  Strengths

  Industry-leading bit cost (wafer cost ~$2,000), lateral scaling technology edge, scale benefits from SanDisk JV, high operating margin.

  Weaknesses

  Heavy customer concentration risk (Apple is largest customer); no exposure to HBM, missing direct participation in HBM’s premium pricing.

  Comparison

  Compared to Samsung and SK hynix, Kioxia holds greater NAND cost control but lacks DRAM/HBM business diversification; versus Micron, Kioxia is more NAND-focused and purer in exposure.

  Risks

  Competitors accelerating NAND capacity expansion post-HBM investment peak; slowdown in AI investment; sustainability of technological edge beyond 332-layer nodes.

Key data

• Target Price¥79,000Based on FY2028 forecast BPS and 4.7x PBR
• Current Share Price¥61,280As of May 28, 2026
• NAND Price Peak TimingQ3 FY2027Expected sequential price increases over next six quarters
• Kioxia Wafer Cost Estimate~$2,000Significantly below Samsung (~$4,000) and Micron/SK hynix (~$6,000)
• FY2028 Operating Profit Forecast¥8.99 trillionSubstantially above consensus of ¥7.47 trillion
• Fair PBR Multiple4.7xBased on 41% average ROE and 8.6% cost of equity

Impact & implications

For Kioxia, this report implies a re-rating of its long-term cost advantage and current market upcycle—creating significant upside potential. For the industry, NAND’s volatility may diminish as AI-driven structural demand provides stability, signaling a broader transition from a purely cyclical commodity to a growth-oriented asset with resilient cash flows. Investors should monitor Kioxia’s expanding share in AI data center SSDs and potential enhancements to its dividend policy.

Risks

• Strong cyclicality in the NAND market—price corrections possible once supply-demand rebalances.
• Slowing AI investment or semiconductor spending constraints reducing demand.
• Uncertainty regarding sustainability of technological advantages—particularly during transitions to higher layer counts.
• Heightened earnings volatility due to high operating leverage during price-up cycles.
• Customer concentration risk (heavy reliance on Apple).

What to watch

• Quarterly sequential NAND price movements—especially whether the peak materializes as expected in Q3 FY2027.
• NAND capex plans of major competitors (Samsung, SK hynix, Micron) following their HBM investment peaks.
• Kioxia’s market share gains in AI data center SSDs.
• Standardization progress and potential 2028 adoption of HBF (High-Bandwidth Flash) technology.
• Specific implementation of Kioxia’s dividend policy and shareholder return initiatives.