Kioxia chief dismisses deeper SK Hynix ties, vows price control

Kioxia CEO Hiroo Ota ruled out a closer manufacturing partnership with SK Hynix, citing antitrust risks and complications from Kioxia's existing joint facilities with Sandisk. Kioxia and Sandisk plan to invest more than ¥5 trillion ($33 billion) to expand Japanese memory production, while SK Hynix is pursuing a 54 trillion won ($40 billion) Korean capacity expansion. Ota said memory prices, which have increased by double- or triple-digit percentages amid AI demand, have risen enough and should not climb further in ways that could damage long-term demand.

Analysis

The relevant equity signal is not a new partnership premium but a more disciplined NAND supply response. SNDK’s earnings sensitivity is asymmetric: incremental wafer additions can support revenue into the next upcycle, but NAND pricing is far more vulnerable than HBM to oversupply because capacity is less specialized and AI-storage demand can be served by competing enterprise SSD architectures. The absence of a broader manufacturing pool reduces near-term antitrust optionality but also limits the probability of an industry-wide capacity cartel.

For SNDK, the key second-order risk is that its capital commitments may arrive just as hyperscalers normalize storage purchases after the initial AI buildout. If NAND contract-price increases outrun cloud capex growth over the next 1-3 quarters, customers will draw down inventories and gross-margin expectations could reset sharply; this is a greater risk to NAND vendors than to HBM-focused 000660 KS. Conversely, constrained joint-fab execution would preserve pricing and favor SNDK through the next earnings cycle.

Consensus may overstate the read-through from AI memory scarcity to NAND. AI infrastructure demand is highly favorable for high-capacity enterprise SSDs, but the marginal economics accrue only if bit-supply growth remains below demand; announced capex is not equivalent to usable output, yet it is a 6-18 month multiple-compression risk. Monitor quarterly NAND bit-growth guidance, enterprise SSD mix, and SNDK gross-margin trajectory rather than headline capacity spend.

Key Decisions for Investors

• Maintain a tactical long bias in SNDK only on pullbacks ahead of the next earnings report; target a 1-3 month hold, with upside contingent on enterprise-SSD mix and gross-margin guidance. Do not chase a partnership-driven move: there is no verified transaction catalyst.
• Use a relative-value expression: long SNDK / short SOXX in equal beta-weighted dollars for 3-6 months. This isolates a NAND pricing-discipline thesis from broader AI-semiconductor valuation risk; exit if SNDK guides NAND bit growth materially above demand or cuts gross-margin outlook.
• For diversified memory exposure, prefer 000660 KS over a pure NAND basket on a 6-12 month horizon: HBM qualification and packaging constraints offer more durable supply protection. The thesis is falsified by HBM customer qualification delays or a material cut in AI accelerator demand.
• Set an alert around the next industry contract-price data and SNDK earnings: two consecutive months of flat-to-down NAND pricing, or a sequential decline in enterprise SSD revenue, should trigger a reassessment of any SNDK long because the operating leverage reverses quickly.