The rapid development of artificial intelligence has the potential to exacerbate global inequalities, warns the UN’s Independent International Science Panel on AI in its preliminary report. The paper was prepared by 40 independent experts, who point out that access to AI tools alone does not guarantee equal benefits for all countries.

The key challenge is the concentration of computing infrastructure and advanced model development. According to the dashboard, the US controls about 75% of the capacity of leading AI supercomputers; China controls about 15%. Together, these two countries provide about 90% of the computing used to train the most powerful systems. Countries that rely on foreign models and cloud infrastructure gain access to technology, but lose control over standards, safeguards, and adaptation to local conditions.

A separate section of the report is devoted to language inequality. Generative models work better with English and other common languages, while rare languages do noticeably worse. Another barrier is basic internet access: 2.2 billion people remained offline in 2025. For these regions, AI adoption begins with energy, connectivity, and digital skills.

The panel does not set binding rules, but it offers directions for public policy: development of local AI infrastructure, increasing AI literacy, investing in local developers, building AI security institutions, and strategies to combat misinformation.