Global AI Trends: US-China Gap Narrows as South Korea Leads in Patents

The performance gap between the United States and China in artificial intelligence has nearly disappeared, according to the 2026 AI Index report from Stanford University’s Human-Centered AI institute (Stanford HAI). Reporting from Chosun Ilbo indicates that U.S. Models now lead in performance scores by only 2.7 percent, with the two nations taking turns occupying top benchmark positions.

While the United States maintains a lead in capital, infrastructure, and AI chips, China has secured an advantage in AI publications, papers, and the development of autonomous robots. Data from February 26, 2026, shows that China holds roughly 60 percent of global AI patents.

Regional Strengths and Patent Trends

The 2026 AI Index report highlights a shift toward global model development, with notable launches appearing in Southeast Asia, Latin America, and the Middle East. However, the concentration of intellectual property remains heavily skewed toward China, which is described as the epicenter of global patenting growth.

South Korea has emerged as a significant player in the intellectual property landscape. According to reports from KBS News and Newsis, South Korea has ranked first in the world for AI patent applications per capita for two consecutive years. The country also ranks third globally for notable AI behind the United States and China.

Despite these patent achievements, South Korean firms face internal challenges. Samil PwC reports that 74 percent of AI performance is concentrated among the top 20 percent of companies. While South Korean enterprises have been fast to implement pilot programs, a gap remains in achieving full automation.

reports from v.daum.net indicate that while South Korea maintains a top position in High Bandwidth Memory (HBM), the domestic AI sector is constrained by limitations in the power grid.

Corporate Transparency and Environmental Costs

The Stanford HAI report identifies a declining trend in transparency as private companies increase their influence over the AI market. Major developers including Google, Anthropic, and OpenAI are no longer disclosing the training time or the scale of training data used for their latest models.

This lack of openness extends to technical documentation; 80 of the 95 major models released in 2025 were published without their training code.

The report also quantifies the environmental impact of large-scale model training and inference. It estimates that the training of xAI’s Grok 4 model emitted approximately 72,000 tons of carbon dioxide. The water required for GPT-4o inference workloads is equivalent to the amount needed for 12 million people to drink.

Global Adoption and Public Trust

Generative AI adoption has reached 53 percent of the global population as regular users, a rate of expansion that exceeds that of smartphones, the internet, and PCs. However, adoption varies significantly by region. Despite its leading role in development, the United States ranks 24th globally in regular use, with a rate of 28.3 percent.

Public trust in AI regulation also shows a wide geographical divide. According to the 2026 AI Index report, 53 percent of citizens in the European Union express trust in AI regulation. In contrast, only 31 percent of U.S. Citizens and 27 percent of Chinese citizens report similar trust.

In the professional sphere, the adoption of AI tools has tripled the individual productivity of scientists. However, the report notes a corresponding decline in research diversity, as the scope of scientific inquiry is increasingly tilting toward topics that are data-rich.