Japan chip earnings roundup: AI a double-edged sword

Japan’s major chipmakers reported mostly positive September-quarter results, with Advantest beating EPS estimates by 27.2% and raising FY2027 operating-profit guidance to ¥967B from ¥846B, while Tokyo Electron also beat and announced a ¥150B buyback. Results and share reactions diverged: Disco missed EPS by 6.9%, Kioxia’s ¥2.0T operating-profit guidance fell short of the ¥2.35T consensus, and Kioxia shares were down 14.63% in the latest session; investors are weighing strong AI demand against moderating memory pricing, valuations, and slower AI spending. A hawkish Bank of Japan and a 2.1% Nikkei decline on Oct. 30 added to pressure.

Analysis

This is a dispersion trade, not a clean signal to own Japanese semiconductors as a basket. AI accelerator complexity supports testing intensity and process-equipment demand, so Advantest and Tokyo Electron have a nearer-term earnings-validation path than memory. But order momentum can reverse before reported earnings if hyperscalers slow capex; the market’s high expectations make guidance and order conversion more important than headline beats. The latest rebound also weakens the entry case for chasing winners immediately.

Kioxia’s risk is more than a quarterly miss: moderating NAND pricing and YMTC’s shipment gains can pressure utilization and pricing across the memory supply chain, while leveraged positioning can amplify both downside and any squeeze. Treat the cited FY2027 revenue outlook as especially sensitive to sustained pricing, not as a dependable base case. Disco’s EPS miss is a caution on earnings conversion even where revenue holds; Renesas’s reported EPS should not be extrapolated without separating recurring operations from one-offs.

Contrarian angle: the reaction may reflect crowded AI expectations and macro/rate sensitivity rather than evidence that AI demand has already rolled over. Conversely, strong equipment guidance is not proof that customers’ capital returns justify current multiples. A hawkish Bank of Japan could add yen-translation and valuation pressure for exporters; company-level currency sensitivities need verification. The key 1–3 month tests are equipment orders/backlog, hyperscaler capex commentary, NAND contract pricing and Kioxia’s next guidance. Over 6–18 months, memory pricing and China competition matter more than this quarter’s AI narrative.

Key Decisions for Investors

• Prefer a staged relative-value position: long Advantest or Tokyo Electron against a smaller Kioxia short, rather than a broad Japan-semiconductor long. Do not add to the equipment leg after sharp rallies without confirmation in orders or customer capex. The pair thesis is falsified by equipment order/backlog deterioration or sustained NAND pricing and Kioxia guidance recovery; broad AI-capex cuts could hurt both legs and widen losses.
• Avoid chasing Kioxia lower after the sharp repricing. Keep it on a tactical short/watch list, and add only if NAND pricing or company guidance weakens further; cover or stand aside if pricing stabilizes and the next outlook improves. Verify balance-sheet leverage and positioning before sizing, since crowded shorts can squeeze.
• For the next 1–3 months, monitor Advantest and Tokyo Electron order/backlog commentary, hyperscaler capex updates, NAND contract prices and Renesas recurring operating results. If equipment demand holds but memory pricing deteriorates, favor equipment over memory; if capex revisions turn negative, reduce the entire semiconductor exposure rather than relying on the pair hedge.