Top 4 Japanese Chipmaking Stocks, According to Bernstein

Bernstein highlighted several Japanese semiconductor stocks as AI infrastructure beneficiaries, with Hoya, Advantest, Kokusai, and Renesas all tied to expanding demand in EUV masks, AI chip testing, DRAM capacity, and data center power semis. Hoya’s mask blanks business is expected to grow at a 17% CAGR over three years, Advantest’s SoC tester TAM has been raised to $8.7B-$9.5B for 2026, and Renesas’ AI-related revenue doubled in 2025 to 4% of sales with another doubling guided for 2026. The piece is broadly positive for the named stocks but is primarily analyst commentary rather than a company-specific catalyst.

Analysis

The cleanest read-through is not “AI capex up,” but that the bottlenecks are shifting from compute demand to manufacturing complexity and test/packaging intensity. That favors the picks-and-shovels with pricing power and installed-base leverage, while the eventual losers are the lowest-value-add parts of the chain where incremental volume can be competed away. In other words, this is a breadth trade in semicap equipment and materials, not just a single-name AI beta trade.

The second-order effect is that AI server rollout is becoming more semiconductor-content intensive per dollar of end demand: more power management, more test cycles, more advanced process steps, more lithography-dependent materials. That creates a virtuous cycle for the named beneficiaries because capacity additions by foundries and DRAM makers tend to arrive in waves, and once those orders are placed they are sticky for multiple quarters. The more interesting implication is that suppliers with “content per unit” tailwinds can outperform even if end-market unit growth slows.

Risk is mostly timing and concentration. The market is likely discounting multi-year adoption curves already, so the near-term risk is a digestion period if AI server orders normalize or if memory capex pauses after an initial surge. The other risk is substitution: if customers redesign around fewer test steps, alternate substrates, or lower-spec power architectures, some of the incremental benefit could leak to competitors not mentioned here.

The contrarian angle is that the move may be underestimating the durability of test and power demand versus pure GPU enthusiasm. These are not one-cycle beneficiaries; they monetize every new node, every package iteration, and every extra rack of AI infrastructure. That makes the setup attractive for a staged entry rather than chasing strength after a sharp rerating.

Key Decisions for Investors

• Go long 6857.T (Advantest) on pullbacks over the next 2-6 weeks; this is the highest-quality direct beneficiary of escalating AI test complexity, with upside from both TAM expansion and multi-year capacity add-ons. Use a 10-15% trailing stop because the stock can de-rate quickly if AI order visibility slips.
• Initiate a long TSM / short global semi-equipment basket pair for 3-12 months: TSM is the demand enabler, while the equipment names are the more crowded expression. This works if wafer demand stays firm but capital intensity disappoints relative to current expectations.
• Buy 6594.T (Nidec) or comparable power-component exposure only on weakness if using as a secondary AI infrastructure play; the cleaner trade is still Renesas, but power semis have a longer runway. Favor call spreads 6-12 months out to limit multiple compression risk.
• For a lower-beta expression, buy Hoya on any post-rally consolidation and hold into the next 12-18 months; the investment case is not next-quarter earnings but durable pricing power in EUV-related materials. Size modestly because the market may already be pricing in part of the long-duration growth.
• Avoid chasing NVIDIA here; use it as a hedge rather than the preferred long. The article’s incremental signal is that second-order beneficiaries may offer better risk/reward than the core GPU complex if AI capex broadens but growth rates moderate.