Nvidia(US:NVDA) 私募排排网·2026-03-25 09:49Core Viewpoint - The article discusses Jensen Huang's "AI Five-Layer Cake" framework presented at NVIDIA GTC 2026, which outlines how value in the AI era is created and distributed across various industries, emphasizing the interconnectedness of the AI ecosystem and its implications for investment logic and asset allocation [3][5]. Group 1: AI Five-Layer Cake Theory - The "AI Five-Layer Cake" consists of five interconnected layers that collectively drive the AI industry's growth, where progress in each layer directly impacts the value realization of the upper layers [6]. - The five layers are: 1. Energy Layer: The foundation of AI, emphasizing the need for efficient energy supply and the projected doubling of global data center electricity consumption to 945 TWh by 2030 [7]. 2. Chip Layer: The core of computational power, with advancements in chip technology critical for AI expansion, including NVIDIA's new GPU architecture expected to achieve 50 PFLOPS [8]. 3. Infrastructure Layer: The physical embodiment of AI capabilities, with significant investments in AI factories and supercomputers, highlighting the importance of cooling technologies and innovative data center designs [9]. 4. Model Layer: The brain of AI, focusing on the transition from language models to physical AI, with open-source models driving demand across the architecture stack [10]. 5. Application Layer: The final interface where AI creates measurable economic value, with a shift towards AI agents capable of executing complex tasks across various sectors [11]. Group 2: Investment Logic from the Five-Layer Cake - Huang's framework provides a comprehensive investment strategy that emphasizes prioritizing foundational layers, driven by the exponential growth of token consumption and the need for heavy asset infrastructure [12][13]. - Key investment logic includes: 1. Bottom-Up Approach: Prioritizing investments in energy, chips, and infrastructure, which are expected to see more stable performance compared to upper layers [14]. 2. Token Economy: The increasing demand for tokens in AI applications, making "cost per token" a critical competitive metric [14]. 3. Heavy Asset Infrastructure: The construction of AI factories and data centers represents a new wave of capital expenditure, akin to a modern infrastructure boom [14]. 4. Positive Feedback Loop: The interdependence of applications, models, infrastructure, chips, and energy creates a strong positive cycle that enhances value across the entire AI ecosystem [14]. Group 3: Layer-Specific Investment Strategies - Energy Layer: Focus on green energy, grid equipment, and storage technologies as core beneficiaries of AI's energy demands [16]. - Chip Layer: Investment in GPUs, LPU, and advanced packaging technologies, driven by domestic alternatives and technological advancements [18]. - Infrastructure Layer: Capitalizing on the construction of AI factories and data centers, with a focus on liquid cooling and optical interconnects [20]. - Model Layer: Targeting investments in general models and open-source ecosystems, while being mindful of competitive pressures [22]. - Application Layer: Emphasizing sectors with high barriers to entry and strong profitability potential, such as embodied intelligence and industry-specific AI applications [24]. Group 4: Overall Industry Outlook - The AI industry is in its early stages of industrialization, with significant long-term growth potential as it transitions from training to inference, driving value across the entire supply chain [26].