Sino-US AI Industry Chain Competition & Investment Insights Based on Five-Layer Cake Theory

From the perspective of Jensen Huang’s AI “Five-Layer Cake” theory, I will provide an in-depth analysis of the competitive landscape of the Sino-US AI industry chain and investment insights for you.

Jensen Huang, CEO of Nvidia, proposed the AI “Five-Layer Cake” theory, which divides the AI industry into five key layers: Energy (Foundation Layer), Chips, Infrastructure, Models, and Applications (Top Layer). This framework reveals the complete value chain of the AI industry and provides a systematic perspective for understanding Sino-US AI competition.

• The U.S. faces a conflict between surging AI computing power demand and insufficient power supply
• Morgan Stanley warns that the U.S. may face a power gap of up to 20% by 2028 [1]
• Microsoft CEO Satya Nadella stated bluntly: “The biggest problem right now is not computing power, but electricity”
• Data centers have become the industry with the fastest-growing power demand, and AI-driven data centers will account for 9% of the U.S.'s total power load [6]

• Jensen Huang pointed out: “In the bottommost energy sector, China has twice the energy of the U.S.” [1]
• The “Eastern Data, Western Computing” project achieves synergy between computing power, energy, and data, with the national intelligent computing total scale reaching 788 EFLOPS
• China’s intelligent computing scale ranks second globally, with eight national hub nodes gathering over 60% of the country’s new computing power

• The Trump administration shifted its chip export policy to China, announcing that Nvidia is allowed to export H200 chips to China but will collect a 25% share [2]
• Nvidia maintains a solid leading position in AI chip manufacturing, with a market capitalization of $4.4 trillion [2]
• The U.S. still retains technological advantages in chip design and manufacturing processes

• China included domestic AI chips in the official procurement list for the first time, with Huawei Ascend, Cambricon, etc., selected [3]
• Domestic chips are scaling up rapidly, developing in parallel with international mainstream chips
• China adopts a dual-track strategy of “domestic chips + international mainstream chips” [4]

• U.S. tech giants tend to build ultra-large-scale data centers, pursuing extreme economies of scale
• Under construction are giant AI computing power factories, with single-project investment scales reaching billions of dollars
• New projects are basically fully liquid-cooled, with 140kW+ cabinets becoming the standard [4]

• China focuses more on optimizing resource allocation nationwide, emphasizing “10,000-card clusters”
• Cabinet power is mainly 60-120kW, with air-liquid hybrid solutions common
• Policies require PUE ≤1.25, promoting the popularization of liquid cooling technology [4]

• Major U.S. tech companies have invested over $350 billion in AI infrastructure, which is expected to exceed $400 billion by 2026 [5]
• OpenAI, Google, etc., maintain leading positions in basic model research and development
• Model training costs are extremely high, requiring computing power-intensive investment

• China’s total investment in AI is close to $100 billion, about 1/3 of the U.S.'s scale [5]
• Companies like DeepSeek have launched new models based on the “Mixture of Experts” architecture, reducing training energy consumption by 70%
• China’s large models have narrowed the gap with the U.S. in multiple tasks

• The U.S. started earlier in AI commercial applications, with a more complete ecosystem
• AI penetration is high in various industries, especially in finance, healthcare, and tech services
• AI startups lead globally in financing scale and quantity

• China’s AI applications are deeper in vertical fields such as manufacturing, urban management, and transportation
• Policies promote deep integration of AI with traditional industries, with diverse application scenarios
• China has advantages in massive data and application scenarios

• AIDC energy storage becomes a new blue ocean: Global new AIDC energy storage installations reached 16.5GWh in 2024, expected to increase to 209.4GWh by 2030, with a compound annual growth rate of 52.7% [6]
• AIDC energy storage demand surges, with companies like CATL and Sungrow Power accelerating layout
• Clean energy investment surges, with wind and solar energy projects entering a golden development period

• Domestic AI chip substitution accelerates, benefiting industrial chains like Huawei Ascend and Cambricon
• Segments such as chip design, packaging and testing, and manufacturing equipment have prominent investment value
• Trump’s policy opening brings short-term opportunities for related enterprises, but long-term independence is still needed

• Liquid cooling technology demand explodes, with cold plate liquid cooling becoming the mainstream solution
• Clear demand for new technologies like 800V power architecture and SST [6]
• The data center construction boom drives performance growth of related equipment manufacturers

• Strong demand for vertical industry large models, with rich opportunities in healthcare, education, finance, etc.
• Model lightweighting and edge computing become new trends
• AI-driven enterprise service software迎来 upgrade and replacement opportunities

• Manufacturing digital transformation accelerates, with broad prospects for industrial AI applications
• Scenario-based applications like smart cities and autonomous driving continue to be implemented
• AI + traditional industry integration creates huge market space

: AIDC energy storage, liquid cooling technology, data center infrastructure
: The explosion of AI computing power demand directly drives infrastructure investment, with high performance certainty for related enterprises

: Domestic chip substitution, vertical industry large models, AI application implementation
: Technological breakthroughs + policy support promote localization, with application scenarios continuously enriching

: AI energy solutions, next-generation computing architecture, AGI basic research
: AI development enters deep water, with energy efficiency and new computing paradigms becoming key

1. Policy Risk
   : U.S. policies toward China’s AI industry remain uncertain, and trade frictions may escalate
2. Technology Risk
   : Breakthroughs in key technologies may change the industry landscape; investors need to maintain technical sensitivity
3. Market Risk
   : AI concept speculation is overheated; need to focus on enterprises’ real fundamentals and profitability
4. Energy Risk
   : Insufficient power supply may restrict AI industry development; energy price fluctuations affect costs

Jensen Huang’s “Five-Layer Cake” theory clearly reveals the multi-layer competitive landscape of the AI industry. Currently, Sino-US competition shows the characteristics of “the U.S. excels in top-layer applications, while China excels in basic energy”. Investors should adopt differentiated investment strategies based on the development stage and competitive landscape of different layers, not only seizing the certain opportunities of current infrastructure construction but also laying out the growth space of future technological breakthroughs and applications.

In this critical period of global AI competition, understanding the underlying logic of the industry and grasping the development rhythm of each layer will be the key to investment success.

---

[1] 21st Century Business Herald - “Turing Award Winner Answers 21: AI Competition Needs to Beware of Power Shortage Risks”
[2] Guancha.cn - “Trump: Allows Nvidia to Export H200 Chips to China, but Takes 25% Cut”
[3] Yahoo Finance - “Trump Approves Nvidia’s Sale of Advanced AI Chips H200 to China”
[4] Sina Finance - “China vs. U.S.: Latest Comparison of AI Data Centers 2025 (Pure Dry Goods)”
[5] Guancha.cn - “When the U.S. Goes All-In on AI, China is Winning Multiple Tech Races”
[6] Securities Times - “The ‘New Blue Ocean’ Under Computing Power Surge: AIDC Energy Storage Track Explodes”