Global Tech 2026: A Dual-Track Strategy Amid AI Super-Cycle Peaks and China’s Rise

Core Conclusion

The global technology cycle in 2026 is characterized by a sharp divergence between robust AI/data center demand and emerging pressures on consumer electronics. We advocate a barbell strategy: maintain Overweight positions in segments with structural pricing power and AI exposure (memory, leading-edge foundry, semicap) while selectively adding exposure to oversold, cash-generative non-AI names. The investment narrative will pivot from a broad-based AI rally in H1 towards managing cost-driven demand destruction and identifying the next bottlenecks in H2.

Evidence Chain

The semiconductor super-cycle continues but growth is bifurcating. Global semiconductor revenue is forecast to reach $1.3tn in 2026, up 68% YoY, with the SOX index expected to deliver 90% EPS growth in H1. Memory and logic remain the primary drivers. However, consensus anticipates a slowdown, implying capital will flow to companies exceeding growth expectations. Non-AI segments, trading at mid-cycle multiples of ~18x NTM P/E, have more upside potential than AI names at ~25x, assuming no new positive catalysts for AI.

Investment Implication: Favor companies with visibility into beating consensus, primarily in memory and semicap. Use market corrections to build positions in high-quality, non-AI analog/consumer semiconductor names.

AI-driven bottlenecks are shifting from logic to memory and front-end wafer capacity. Memory remains a critical constraint for AI inference, with DRAM pricing moving past all-time highs and a severe HBM supply deficit (only 2% sufficiency estimated for 2026). This capacity-constrained cycle, with unprecedented 2-3 year order visibility, is driving memory capex. The next bottleneck is likely extreme ultraviolet (EUV) lithography equipment, setting the stage for a multi-year semicap upcycle with wafer fab equipment (WFE) growth accelerating to 25%+ in 2026-27 from 16% in 2022-25.

Investment Implication: Overweight memory producers (Samsung, SK Hynix, Micron) and semicap leaders levered to DRAM and advanced logic (ASML, AMAT, Advantest).

China’s semiconductor ecosystem presents a structural, decoupling-driven opportunity. China's AI chip self-sufficiency is projected to rise from 33% in 2024 to 76% by 2030, creating a parallel $67bn market by 2030. Domestic innovation is focused on lower-cost inferencing and physical AI, driven by sovereign and SOE demand. This supports localized manufacturing and equipment beneficiaries, independent of the global cycle.

Investment Implication: Overweight select China semiconductor manufacturers (SMIC), domestic equipment champions (NAURA, AMEC), and companies enabling the local GPU supply chain.

Key Debates and Risks

• Tech Inflation & Demand Destruction: Rising costs for wafers, OSAT, and memory will pressure margins for chip designers and consumer electronics (PCs, smartphones) in H2 2026, testing price elasticity.
• AI Investment Payoff Uncertainty: The bull case requires visible productivity gains and revenue growth from AI, not just cost-cutting. The bear case foresees potential indigestion from over-investment and unclear return paths, compounded by power and funding constraints.
• Cycle Timing and Magnitude: The current cycle is supply-driven in memory. A mismatch between soaring contract prices and subdued spot prices signals potential inventory building, raising the risk of a sharper correction if end-demand softens.
• Geopolitical & Regulatory Execution: The decoupling of AI compute value chains creates opportunity but also persistent volatility. Success for China’s domestic players hinges on flawless technological execution amid ongoing trade restrictions.

Valuation and Trade Implications

We maintain a barbell approach within a broadly constructive semiconductor stance.

• Overweight: Memory (005930 KS, 000660 KS), Foundry (2330 TT), Semicap (ASML NA, AMAT US, 6857 JT).
• Selective Exposure: China semis localization plays (0981 HK, 002371 CH).
• Underweight/Caution: Consumer electronics-exposed names, legacy foundry, and IC designers with high exposure to rising input costs.