诺安基金邓心怡：AI产业内部正在发生深刻的结构性演变

Group 1: Core Expectations in AI Industry - The AI industry is currently focused on three core expectation divergence directions: application, computing power, and core consumer terminals related to human interaction [1][12] - The industry is undergoing profound structural changes, although the framework for judgment remains unchanged [1][12] Group 2: Computing Power - The computing power industry has transitioned from "demand validation" to "profit validation" and "structural differentiation" [2][13] - The evolution of stages includes: 1. Demand validation: Concerns about the authenticity of AI applications were disproven around September 2023 [3][14] 2. Revenue validation: Fears of AI falling into a "burn rate trap" have diminished with rapid growth in annual recurring revenue (ARR) from companies like OpenAI [3][14] 3. Profit validation: Current market concerns focus on whether AI can generate sustainable profits, as evidenced by the divergence in operating profit margins reported by cloud service providers and internet giants [3][15] - A core contradiction exists between rigid costs and elastic revenues, exemplified by OpenAI's high fixed costs and the challenge of maintaining revenue growth [4][16] - Google's "full-stack closed-loop" ecosystem demonstrates strong cost control and business model resilience, leading to a reevaluation of its market value [5][17] - The structure of computing power investment is evolving towards a "dual-track" ecosystem, moving from Nvidia's dominance to a parallel structure with Google TPU and other ASICs [5][17] - Demand for computing power is shifting from training to inference, providing opportunities for second-tier suppliers and new technologies to emerge [6][18] Group 3: Applications - The evolution of data in AI is entering a "research era," highlighting the value of vertical data [7][19] - The three stages of data evolution include: 1. Public internet data pre-training has reached a bottleneck [8][20] 2. Reinforcement learning and synthetic data enhance specific model capabilities but lack comprehensive judgment [8][20] 3. The next breakthrough requires embedding complex value judgment systems into models, moving beyond simple pre-training [8][20] - Investment opportunities are shifting towards companies that can effectively utilize vertical data and integrate AI into business processes to enhance efficiency and create commercial value [9][20] Group 4: Consumer Terminals - The transformation of consumer electronics, as the final interface for AI-human interaction, presents significant investment opportunities [10][21] - Future breakthroughs will depend on devices' ability to continuously collect and process real-time, high-frequency, unstructured data to provide personalized recommendations [10][21] - Investment logic suggests a focus on two types of companies: 1. Strong beta companies closely partnered with leading model companies like Google and Android [10][22] 2. Innovators of key components that will play a crucial role in new interaction forms, moving beyond traditional screens [10][22]