The following insights are excerpted from Noah Fund manager Deng Xinyi's latest research notes:
Three core areas of expectation gaps in AI remain under close watch: applications, computing power, and consumer-facing terminal devices. While this framework remains unchanged, the industry is undergoing profound structural evolution.
1. Computing Power: Transitioning from "Demand Validation" to "Profit Verification" and "Structural Differentiation" The overseas computing power supply chain remains AI's primary beta driver, with its long-term demand unquestioned. However, the industry has entered its third phase, shifting focus from "demand authenticity" to "profit realization."
Phase Evolution Recap: - Phase 1 (Demand Validation): Market skepticism about AI applications' viability was dispelled by September 2023 when sustained token usage growth was confirmed. - Phase 2 (Revenue Validation): Concerns about AI's monetization capability faded as companies like OpenAI demonstrated rapid ARR growth through funding disclosures. - Phase 3 (Profit Verification): Current market anxiety centers on sustainable profitability, evidenced by diverging operating margins in recent CSP and tech giant earnings reports.
Core Dilemma: Rigid Costs vs. Elastic Revenue OpenAI exemplifies this challenge with its inflexible cost structure - comprising top-tier talent compensation and NVIDIA's non-negotiable GPU procurement. Despite cost-saving measures like equity swaps and revenue diversification through ChatGPT subscriptions, expenses continue outpacing income growth, prompting industry-wide reassessment of business sustainability.
Structural Shifts: Google's "Full-Stack" Model and Dual-Track Computing Ecosystem Google's vertically integrated approach (TPU infrastructure + Gemini models + search/cloud/Android ecosystem) demonstrates superior cost control and business resilience. Its successful Gemini 3 deployment via proprietary TPUs signals a broader trend where major players will adopt diversified computing solutions to reduce vendor dependency. This drives two key transformations:
- Chip Level: Transition from NVIDIA's GPU dominance to a dual-track system incorporating ASIC solutions like Google's TPUs, explaining Broadcom's periodic outperformance as Google's key supplier. - Cluster Demand: Shift from training-focused to inference-dominant workloads, relaxing extreme yield requirements for components like optical modules and creating opportunities for secondary suppliers. Emerging technologies like optical circuit switches (OCS) will address inference-specific needs for low latency and power efficiency.
2. Applications: Data Evolution Enters "Research Era" with Vertical Data Rising Model capability advancement faces data source bottlenecks, marking AI's transition into a "research era" of data utilization.
Three Data Evolution Stages: - Public Internet Data Pretraining: Now reaching saturation. - Reinforcement Learning/Synthetic Data: Produces specialized but imbalanced capabilities. - Value Judgment Systems: Next breakthrough requires embedding sophisticated, integrated value assessment frameworks.
Investment Opportunity: With public data dividends exhausted, value creation will emerge from effectively leveraging proprietary vertical data to enhance productivity. Enterprises combining unique industry data with deep AI integration will realize tangible business value.
3. Consumer Terminals: Frontier for "Thematic Investments" in New Product Categories As AI's primary human interface, consumer electronics represent high-potential thematic investments despite longer development cycles.
Key Driver: Future breakthroughs require devices capable of continuously processing real-time, unstructured life data to deliver personalized recommendations. Early experiments demonstrate this potential through wearable-powered life coaching via local AI models.
Investment Thesis: This represents radical product innovation rather than incremental upgrades. Traditional electronics giants may not lead this transition due to legacy business constraints. Focus areas include: - High-Beta Plays: Ecosystem partners closely aligned with leading model developers (e.g., Alphabet/Android). - Component Innovators: Companies mastering new interaction hardware, potentially replacing traditional screens with projection engines or novel sensors in the AI era.
Comments