China AI 2.0: Shifting from Capability to Value Capture – A New Investment Phase

Core Thesis

The China AI narrative has fundamentally shifted from a technology catch-up race to a value capture phase. The market is pricing the upfront infrastructure buildout (power, semiconductors, hardware) but underestimating the accelerating “rate of change” in commercial adoption and earnings conversion. Over the next 12–18 months, the most incremental returns will migrate from enablers to adopters—companies embedding AI into products and services to generate measurable margin expansion and EPS growth.

Evidence Chain: Faster Adoption, Tangible Profit Impact

1. Adoption velocity is accelerating across the enterprise. The share of companies that are either AI enablers or adopters has risen from ~43% to ~51% over the past two years. Critically, 47% of CIOs now plan to launch their first AI project within 12 months, up from 40% six months ago. This is not a vague pipeline—it is a near-term spending catalyst.

2. Financial outcomes are already visible. Over the past two years, forward 12-month EPS for identified AI adopters has risen ~62%, versus just 10% for the broader MSCI China Index. EBIT margins for this group are projected to expand by 12–13 percentage points to 16–17% by 2027, far outpacing the market average. The driver is operational leverage from AI-driven efficiency, not just top-line growth.

3. Macro productivity gains are structural but gradual. AI is projected to lift China’s total factor productivity cumulatively by ~3 percentage points over the next decade, pushing 2035 potential GDP ~3.5% above the no-AI baseline. However, near-term growth impact will be muted as AI-related labor displacement (especially in white-collar and service roles) offsets early efficiency gains. This creates a mid-cycle buying opportunity for investors with a 2–3 year horizon.

4. Semiconductor self-sufficiency lowers deployment cost risk. Domestic chip capability is expected to rise from 41% in 2025 to 86% by 2030, enabling cost-competitive inference at scale. This structural supply shift supports our above-consensus adoption projections.

Key Divergences & Risks

Market misjudges the “rate of change.” Infrastructure plays (power, semi, hardware) have outperformed on early capex cycles. But the next leg belongs to application-layer companies where AI revenue and margin inflection are just beginning. The market is under-appreciating how quickly cost savings translate into EPS upgrades.

Risk #1: Enterprise monetization is still early and lumpy. Many B2B AI use cases remain cost-saving rather than revenue-generating. Revenue ramp may disappoint if macro weakness delays IT budgets.

Risk #2: Computing bottlenecks persist. Despite progress in self-sufficiency, access to advanced process nodes, EDA tools, and high-bandwidth memory remains constrained. A supply-side shock (e.g., further export controls) could slow deployment timelines.

Risk #3: Labor disruption could weigh on domestic demand. AI’s displacement of white-collar and service jobs may exacerbate youth unemployment and weigh on consumption, partially offsetting productivity gains. Policy responses (e.g., subsidies for labor-intensive sectors) can mitigate but not eliminate this drag.

Investment Implications & Positioning

Current phase favors enablers, next phase favors adopters. The core AI enablers—foundational model companies (MiniMax, Zhipu) and full-stack platform Alibaba—remain key holdings for structural exposure. For near-term alpha, we shift focus to application-layer names with proven AI monetization:

• Beisen Holdings (9669.HK), Meitu (1357.HK), Roborock (688169.SS), Midea (000333.SZ), Ecovacs (603486.SS) – these companies show favorable risk/reward as AI adoption drives margin expansion and EPS upgrades.
• In power infrastructure, the bottleneck has shifted from compute to electricity. Key beneficiaries: CATL (300750.SZ/3750.HK), Anhui Yingliu (603308.SS), Sieyuan Electric (002028.SZ).
• Semiconductor localization remains a long-term secular play: Cambricon (688256.SS), Iluvatar CoreX (9903.HK), NAURA (002371.SZ), AMEC (688012.SS), ACMR (ACMR.O), SMIC (0981.HK), Unimicron (3037.TW).

Valuation context: The application group trades at an average forward P/E of ~18–22x, with EPS growth accelerating to 25%+ over the next 2 years. That represents a PEG ratio below 1.0x for several names—significantly cheaper than global AI comps. The re-rating catalyst is delivery of 2026–2027 margin guidance.

Position sizing: Overweight adopters relative to enablers, with 10–15% allocation to power infrastructure as a core structural hedge against compute-to-power resource bottlenecks. Monitor labor market data and enterprise AI spending surveys for risk signals.

---

Appendix: Key AI Adopters Under Coverage (selected)