Jensen Huang's 'Five-Layer Cake' Theory and Analysis of Sino-US AI Competition Dynamics

1. Bottom Layer: Energy and Computing Power Guarantee
   
   Jensen Huang positions energy as the foundation of the AI industry and points out that ‘without energy, you cannot build chip factories, supercomputer factories, or data centers’. China’s current power supply (especially the regional supporting facilities combining hydropower, wind power, and solar energy) is twice that of the United States, enabling China to deploy large-scale computing power parks at lower cost and in an extremely short cycle (at the speed of ‘building a hospital in one weekend’)[1]. In contrast, the expansion of U.S. data centers is limited by aging power grids, slow approval processes, and high electricity prices; many proposed centers are even left vacant due to lack of power supply[1]. Additionally, Morgan Stanley warns that the U.S. may face a power gap of up to 20% by 2028, meaning AI chips could be idle without electricity[2].
   
   Investment Tip
   : Prioritize energy infrastructure companies with a ‘computing power as a service’ ecosystem (such as high-efficiency power management, regional microgrids, new energy + energy storage integration systems) and supporting builders participating in ‘East-West Computing Network’-style projects, as they will benefit long-term from computing power concentration and green transformation needs.

2. Second Layer: Chips
   
   The U.S. still leads in high-end AI chip design, but Jensen Huang reminds us: ‘Semiconductors are a manufacturing process’, and China has room to catch up in manufacturing and cost-effectiveness. By 2025, Chinese companies have demonstrated the ability to ‘complete the same task with fewer GPUs’ in large model inference cards and systems; domestic AI chips (e.g., Kunlun Chip) rank among the top globally in sales in several niche markets[3].
   
   Investment Tip
   : It is recommended to layout (1) high-end Taiwanese/European and American design companies integrated with Chinese packaging and testing firms, (2) domestic AI chip + packaging and testing ecosystems (e.g., SMIC, Changjiang Electronics Technology, etc.) to hedge against the risk of Sino-US technology chain differentiation; also pay attention to global supply chain substitution needs under the background of sanctions.

3. Third Layer: Infrastructure (Data Centers, Networks, Scheduling)
   
   China’s infrastructure response speed is significantly ahead, while the U.S. lags due to multiple bottlenecks such as approval cycles, land acquisition, and power access, leading to a substantial gap in AI data center deployment rhythms between China and the U.S.[1]. China’s ‘East-West Computing Network’ and ‘Central Computing Hub’ strategies, through tripartite collaboration between government, operators, and cloud providers, form a ‘government paves the way, enterprises operate’ promotion model, which has initially released 80% of the country’s computing power capacity[3].
   
   Investment Tip
   : Focus on cloud service providers with cross-regional self-building/operation capabilities (e.g., Alibaba Cloud, Huawei Cloud, Tencent Cloud) and power and cooling equipment suppliers centered on AI data centers; also consider U.S.-local high-efficiency data center REITs or operators accelerating digital transformation to obtain long-term rental/operation income for ‘constructing integrated AI infrastructure’.

4. Fourth Layer: Models (Large Models + Open Source Ecosystem)
   
   Jensen Huang emphasizes that China has an open source advantage at the model level; open source promotes low-threshold entrepreneurship and academic innovation, allowing universities, public research institutions, and industry partners to share basic models. China has nearly half of the world’s AI researchers, over 70% of AI patents come from China, and a large number of open source frameworks (e.g., self-developed LLMs, model compression tools) are filling the Western closed ecosystem[1].
   
   Investment Tip
   : It is recommended to关注 large model training and inference platforms (e.g., Zhipu AI, Wanjuan, MiniMax) and ‘basic model + industry adaptation’ vendors; prioritize configuring SaaS/platform companies that can provide model optimization, knowledge distillation, and edge deployment toolchains.

5. Fifth Layer: Applications (Software-based Implementation)
   
   Jensen Huang believes that ‘whoever can apply AI the fastest and most widely will win’. With demographic dividends and policy support, China is exploring multiple implementation paths in smart manufacturing, smart cities, healthcare, government services, and other fields, forming a ‘AI + industry’ composite ecosystem[3]. In contrast, the U.S. focuses more on general models and then outputs them to enterprises through a few large platforms.
   
   Investment Tip
   : Focus on AI application companies with deep industry penetration and data closed loops (smart government, financial risk control, manufacturing digital twins, etc.) and SaaS vendors providing application-level LLMs, which can achieve ‘model value conversion’ faster.

• High-end Manufacturing (Energy, Chips, Infrastructure)
  : Prioritize ‘hard technology’ configuration. Among the five layers, these three layers involve capital intensity, long cycles, and obvious policy support. Entry can be made through leading companies in industrial chains such as energy infrastructure, semiconductor manufacturing, and data centers; risks lie in uncertainties brought by policies and cross-border export controls, so regional (e.g., Singapore/Southeast Asia/China’s inland) and supply chain diversification strategies should be combined.

• Models + Applications
  : Lean towards software logic; optional AI software services with higher growth potential and enterprise-level implementation, including training platforms, industry LLMs, AIops, smart manufacturing SaaS, etc. Focus on ‘model capabilities + industry data’ and ‘profit paths’ (e.g., subscription + scenario revenue). Since the U.S. still has advantages in model and ecosystem control, you can allocate U.S. stocks (e.g., NVIDIA) to maintain exposure—although its market value/growth level is high, its AI data center-related revenue (accounting for over 88% of data center revenue) reflects industrial value conversion[0].

• Hedging Strategy
  : In Sino-US competition, it is recommended to configure a ‘structural dual core’: one is localized assets relying on China’s energy and infrastructure advantages; the other is betting on U.S. AI computing and model commercialization leaders (e.g., NVIDIA) to capture value from technological leadership and global expansion. Pay attention to the impact of exchange rates, regulation, and export controls.

Jensen Huang’s ‘Five-Layer Cake’ not only describes the industrial path but also clarifies investment stratification: the bottom and middle layers are capital-intensive, have long cycles, and are clearly supported by policies. Entry can be made through leading companies in industrial chains such as energy infrastructure, semiconductor manufacturing, and data centers; risks lie in uncertainties brought by policies and cross-border export controls, so regional (e.g., Singapore/Southeast Asia/China’s inland) and supply chain diversification strategies should be combined.

[0] Jinling API Data: NVIDIA Company Overview (December 16, 2025).
[1] Sina Finance: ‘It Takes Three Years to Build a Data Center in the U.S., But China Can Build a Hospital in One Weekend’ (December 7, 2025) https://finance.sina.com.cn/roll/2025-12-07/doc-infzyswf2602778.shtml
[2] 21st Century Business Herald: ‘Turing Award Winner Answers 21: AI Competition Needs to Alert to Power Shortage Risks’ (December 9, 2025) https://www.21jingji.com/article/20251209/herald/d25415c20868ca3f8d6ad347de5199f9.html
[3] Yahoo Finance Hong Kong: 'Burning 3 Trillion Pounds in Five Years Still Not Enough! Enterprises Borrow Money to Rent Computing Power, Tech Giants Spend Heavily to Build Cycles—Experts: Computing Power Has Become the