01 The Historical Context: A Fundamental Productivity Revolution

Haoyu Wang defines the current AI wave not as a short-term trend, but as a "General Purpose Technology" revolution comparable to electricity, the steam engine, or the microprocessor.

• The 80-Year Accumulation: Today’s explosion is the result of eight decades of research reaching a "realization phase" through the convergence of computing power, algorithms, and data.

• Early Stages: Compared to the history of computing, current flagship AI products are akin to the "DOS system" era—historically significant but functionally primitive relative to their future potential.

02 The Core Variable: The Collapse of "Unit Cost of Intelligence"

The most critical driver of the AI era is the rapid decline in the cost of intelligence.

• From Luxury to Commodity: Just as bandwidth and storage became cheap resources, "intelligence" is transitioning from a high-priced asset to a ubiquitous utility.

• Moore’s Law Outpaced: The cost of LLM inference and tokens is falling faster than traditional semiconductor Moore's Law. This "cost revaluation" is the prerequisite for large-scale industrial adoption.

• Shortage vs. Expansion: Current GPU shortages are a classic early-stage infrastructure bottleneck. Massive capital expenditures by global tech giants will eventually turn computing power from a scarcity into a commodity.

03 The Ecosystem: A "Pyramid" Structure

Wang envisions a multi-layered AI ecosystem rather than a "winner-take-all" model:

• The Apex: A few ultra-large models handling the most complex, general tasks.

• The Base: A massive foundation of open-source, small, vertical, and edge models. These solve high-frequency, specific problems with superior cost-efficiency.

• The Commercial Thickness: The "base" of the pyramid often holds more commercial depth and sustainable profit pools than the "apex."

04 Global Landscape: The U.S.-China Bipolarity

The global AI race has consolidated into a systemic competition between two core hubs:

• The U.S. Edge: Leadership in frontier research, advanced chips, and mature capital markets.

• The China Edge: Strengths in engineering implementation, massive data application scenarios, manufacturing ecosystems, and rapid cost reduction.

• The "Execution" Factor: China’s ability to quickly lower costs and deploy products across a complete supply chain makes it a formidable competitor in the commercialization phase.

05 Business Model Shift: From "Selling Resources" to "Selling Results"

As the cost of intelligence drops, the pricing logic of AI will transform:

• Value-Based Pricing: Companies will move away from charging per token or per compute cycle toward charging for measurable business outcomes (e.g., time saved, revenue growth, or risk reduction).

• Vertical Workflow Integration: The true "moat" for AI applications is not the model itself, but the ability to reconstruct specific industry workflows.

06 Asset Allocation: Infrastructure, Applications, and Macro-Synergies

Merit Asset Management follows a multi-layered allocation strategy:

1. Infrastructure (The "Shovel Sellers"): Compute chips, servers, data centers, and power equipment. These assets lead the expansion phase.

2. Applications & Platforms (The "Gold Miners"): Companies that embed AI into vertical workflows to generate repeatable, high-margin revenue.

3. Macro Beneficiaries: Physical assets that support the digital expansion, such as copper, power grids, and high-end manufacturing.

Core Conclusion:

The ultimate winners of the AI revolution will not necessarily be the "smartest" models, but the entities that can most effectively redistribute industrial profits once intelligence becomes a low-cost, foundational resource. Investors should focus on the transition from Model Height to Application Thickness.