Closing the Gap? Measuring and Explaining US-China Differences in LLM Performance

Abstract

This thesis contributes to the ongoing debate over global AI leadership by providing a quantitative assessment of the performance gap between the United States and Chinese large language models (LLMs) and the macroeconomic factors underlying this gap. As artificial intelligence becomes increasingly embedded into society, from economic to military domains, understanding which technology ecosystem, Western or Eastern, will shape its future is becoming increasingly important. To address this question, the study constructs a Composite Performance Index (CPI) to provide a standardized measure of model performance across multiple benchmarks and domains, enabling direct cross-country comparison. Using this framework, the analysis evaluates both frontier LLM performance and average performance trends over time. The results show that while the United States maintained an early lead in frontier LLM development, Chinese models rapidly converged and closed the frontier gap by mid-2025. At the same time, the average performance of Chinese models overtook that of US models by mid-2024, suggesting a broader convergence of the LLM ecosystem. However, following this period of convergence, Chinese frontier model performance shows signs of plateauing, which is attributed to a shift in national AI priorities alongside ongoing restrictions on access to advanced semiconductors. The study finds that access to advanced semiconductor chips is the primary constraining factor on frontier performance, while other inputs, including energy, investment, and human capital, function as enabling factors rather than binding constraints. Moreover, the findings suggest that differences in policy priorities and institutional strategies, rather than resource availability alone, play a central role in shaping the trajectory of LLM development. The thesis is structured as follows. Chapter 1 introduces the research question and key hypotheses. Chapter 2 provides background on LLM development and differences between the US and Chinese LLM ecosystem. Chapter 3 outlines the conceptual framework and methodology of both parts of this study. Chapter 4 presents the construction and measurement of the US-China LLM performance gap, while Chapter 5 evaluates its underlying drivers. Finally, Chapter 6 concludes with a summary of findings and policy implications.