AlphaLens
Research
研报5月10日 · Morgan Stanley

Asia Technology: China AI 2.0 – Entering a New Phase

中文EN⚠ quality lint: see notes

Asia Technology: China AI 2.0 – Entering a New Phase

Core Conclusion

China AI 2.0 shifts from capability catch-up to value capture. The narrative moves from training to inference, technology to application, and potential to monetization. AI adopters now show 62% EPS growth (vs. MSCI China 10%) and EBIT margins expanding from 4% (2021) to 16-17% (2027E). The rate of change in AI materiality drives outperformance: companies with rising AI importance beat MSCI China by 88ppt in the past year. Structural opportunities cluster in power storage, humanoid robotics, autonomous driving, and full-stack AI platforms. The market underappreciates margin expansion scale, domestic chip TCO advantage (30-60% below Nvidia China processors), and AI cloud CAGR of 72%.

Evidence Chain

1. AI Adopter Profitability Is Inflecting Adopters (Beisen, Meitu, Roborock, Midea, Ecovacs) show NTM EPS up 62% over two years vs. 10% for MSCI China. EBIT margins to reach 16-17% by 2027E from ~4% in 2021. 91% of adoption benefits come from cost efficiency, not revenue. This margin expansion is structural: AI enables headcount leverage and process automation in software, manufacturing, and consumer services.

2. Adoption Breadth and ‘Rate of Change’ Drive Returns AI adopters and enablers rose from 31% of coverage to 37% over two years. The basket of companies with rising AI importance outperformed MSCI China by 88ppt. Sectors with highest rate of change: utilities (+44% exposure), industrials (+16% importance, driven by embodied AI supply chain), and consumer discretionary (+9% importance, +11% exposure). Investment implication: focus on companies where AI materiality is accelerating, not just current exposure.

3. Domestic AI Chip Self-Sufficiency Improves Cost Structure Local semiconductor self-sufficiency projected from 41% (2025) to 86% (2030). Domestic AI accelerator TCO is 30-60% lower than Nvidia China processors, with comparable per-token cost on inference workloads. SMIC N+2 capacity to grow from 22k wpm (2025) to 51k wpm (2027). This enables scaled, cost-effective AI deployment independent of US export controls. Key beneficiaries: Cambricon, Iluvatar, MetaX, SMIC.

4. AI-Powered Energy Storage Enters Inflection Global ESS annual incremental deployments CAGR 30% to 2030, with data center demand reaching 85GWh (China) and 169GWh (US) by 2030. Yingliu and Sieyuan show 2025-27E earnings CAGR of 40-70%, PEG <1.0x vs. global peers 1.7x. Each 1ppt global market share gain adds 21-31% upside to 2027 EPS. Shift from power availability to power flexibility creates structural demand.

5. AI Cloud Market and LLM Pricing Power Emerge China AI cloud TAM to grow from Rmb15bn (2024) to Rmb218bn (2029), 72% CAGR. Frontier LLM API input prices rose 80% from 2Q25 to 1Q27; output prices +36%. MiniMax ARR US$150mn (Feb 2026), Z.ai ARR US$250mn (Mar 2026). Hyperscalers regain share from SOEs; ByteDance disrupts with 15% AI IaaS share. Alibaba (SOTP US$245, cloud contributing US$91) is best full-stack AI platform; Tencent best application-layer.

Key Risks

  • Labor displacement near-term: AI substitution hits white-collar, middle-income roles first. Youth unemployment already in high-teens. Policy may protect labor-intensive services but allow faster replacement in knowledge roles, creating macro drag before productivity gains.
  • Monetization imbalance: 91% of adoption benefits are cost efficiency vs. 9% revenue. Enterprise AI monetization (especially 2B) is early and uneven. Software vendors face structural disruption from agentic AI.
  • Semiconductor bottlenecks: Advanced lithography, EDA tools and manufacturing equipment remain constrained. SMIC N+2 is limited; some chips tape out on N+1 with lower density.
  • Data gaps and framework bias: Some markets lack transparency. Current framework may overweigh near-term winners and underweight earlier-stage opportunities (e.g., humanoid robotics).
  • Policy and macro uncertainty: US technology restrictions, infrastructure availability, and regulatory changes could alter competitive dynamics. The 15th FYP signals caution on job disruptions but may accelerate diffusion in high-end roles.

Valuation and Trading Implications

Current valuation supports overweight on AI adopters with visible margin expansion (Beisen, Meitu, Roborock, Midea, Ecovacs). Among enablers, Alibaba (OW, SOTP US$245) captures full-stack advantage; Tencent (OW) leads application monetization. In power, Yingliu and Sieyuan trade at PEG near 1.0x vs. 1.7x global peers, offering upside with 40-70% earnings CAGR. In humanoid robotics, Geekplus and emerging players are early-stage but high-growth. The “rate of change” basket remains the primary performance driver: focus on companies whose AI materiality or exposure is rising fastest, not static exposure.