Research
研报5月4日 · Morgan Stanley

Greater China Semiconductors: China CIO survey on cloud, old memory and edge AI

China CIO Survey Signals Strong Cloud and Edge AI Capex, but Memory Strategy Diverges — OW on Aspeed, Montage, Espressif

Core Conclusion

The 1H26 China CIO survey points to accelerating public cloud adoption and rising physical AI budgets, reinforcing an Overweight stance on cloud semiconductor plays (Aspeed, Montage). However, CIOs are overwhelmingly adopting cost-mitigation tactics for memory rather than expanding budgets, capping upside for old memory (DDR4) demand despite continued firmness. Edge AI exposure also supports an Overweight on Espressif.

Theme 1: Public Cloud and AI Capex Momentum Accelerates

Conclusion: China CIOs are prioritizing AI and data center build-out at the highest levels in survey history, directly supporting cloud semiconductor demand.

Evidence: AI/ML/PA ranks as the #1 spending priority, with data center build-out at #3. GenAI deployment on the public cloud over the next 12 months reached a survey high of 54%, up from 44% in 2H25 and 28% in 1H25. Application workloads running on public cloud reached 31%, up modestly from 30% in 2H25 and 24% in 1H24. Physical AI investment allocation rose to 63% of CIOs from 57% in 2H25, and the average share of IT spending on physical AI is set to increase to 5.7% from 3.8%.

Investment Implication: This sustained cyclical upswing in cloud capex supports revenue and earnings growth for cloud infrastructure suppliers. Aspeed and Montage are direct beneficiaries.

Theme 2: CIO Memory Procurement Behavior Signals Capped Upside for Old Memory

Conclusion: CIOs are responding to memory price inflation primarily through demand management, not budget expansion, limiting volume upside for commodity memory (DDR4).

Evidence: Only 20% of CIOs plan to increase hardware budgets outright, and only 10% see no impact from memory prices. The remaining ~70% are adopting strategies to reduce memory procurement without adjusting budgets: 38% will right-size instead of over-provisioning; 17% will buy only mission-critical capacity and defer non-critical upgrades; 13% will delay purchases for future vendor concessions; 10% will use software optimizations; 8% will renegotiate or use LTAs; 8% will buy de-spec’ed products. This implies that current DDR4 pricing has become unaffordable for some customers, with DDR4 demand remaining firm but growth capped.

Investment Implication: Old memory suppliers face a volume ceiling. Remain selective. DDR4 demand is not collapsing, but the survey confirms a risk of price elasticity constraints that limit upside to volume growth.

Theme 3: Edge AI as a Structural Growth Vector

Conclusion: Rising physical AI budgets are a structural tailwind for edge AI semiconductor names, primarily Espressif.

Evidence: The proportion of CIOs allocating budgets to physical AI increased six percentage points from 2H25. The average share of IT spending on physical AI is forecast to rise nearly 50% to 5.7% from 3.8%. This suggests a broadening of AI investment beyond cloud data centers into edge and endpoint applications.

Investment Implication: Espressif’s exposure to edge AI via MCU and connectivity solutions positions it to capture this incremental spend, supporting an Overweight rating despite near-term cyclical noise.

Key Risks

  • Downside risk from weaker cloud demand: Should macroeconomic conditions deteriorate, CIOs may still defer or cut cloud and AI budgets, particularly if corporate profitability weakens.
  • Slower technology migration or competition in cloud semis: For Aspeed and Montage, delays in spec migration to next-gen interfaces or heightened competition from US or domestic peers could compress margins and growth rates.
  • Margin erosion in edge AI segment: For Espressif, more intense competition from local MCU players or slower-than-expected adoption of edge AI use cases could pressure margins.
  • Policy risk in China: Further US-China tech export controls or new domestic regulatory tightening could disrupt supply chains or limit market access, particularly for Montage and Aspeed.

Appendix: Key Survey Data Compression

MetricCurrent (1H26)Prior (2H25)Change
CIOs allocating IT budgets to physical AI63%57%+6pp
Avg share of IT spending on physical AI5.7%3.8%+1.9pp
GenAI deployment on public cloud intent (12m)54%44%+10pp
Application workloads on public cloud31%30%+1pp
CIOs increasing hardware budgets outright20%N/AN/A
CIOs using cost-mitigation strategies vs budget expansion~70%N/AN/A

Related (同 ticker)