Wireva

Few Countries Control Global AI Power as US and China Dominate

The United States and China hold most of the computing power, energy, research capacity and data behind advanced artificial intelligence, leaving most other countries with little say in how the technology is built and governed.

This item was produced with AI assistance under the editorial responsibility of Haydamax OÜ.

The United States and China hold most of the power to shape artificial intelligence, the technology many now compare to electricity and to the industrial revolution of this century. The United States has long led the field, but China's rapid rise has made it the strongest challenger to American dominance. The result is an increasingly bipolar AI race in which most of the rest of the world depends on decisions made in Washington and Beijing.

The concentration is visible first in computing infrastructure. Between 2022 and June 2026, the United States received roughly 72% of the world's new AI computing capacity, according to estimates by Epoch AI, a research organization. China received 10%, leaving just 18% for all other countries combined. The gap extends to energy. The International Energy Agency estimates that the two countries accounted for about 70% of the electricity consumed by data centers worldwide in 2024, while data centers in the rest of the world consumed about 30%.

Research capacity follows the same pattern. A Carnegie Endowment study of more than 10,000 researchers who published at NeurIPS, a leading AI conference, in 2025 found that three-quarters were affiliated with employers in the United States or China. Those institutions shape which problems get studied, which systems get tested and which safeguards get prioritized, and most countries have little role in those choices even as the technology spreads into schools, hospitals and government offices around the world.

Language and cultural representation raise a separate set of concerns. Most of the world's roughly 7,000 languages remain poorly supported by AI, although about 3.4 billion people speak them. The data used to train frontier models and the benchmarks used to evaluate them largely leave these languages out. A recent snapshot of Common Crawl, a web archive widely used to train language models, found that roughly 43% of its pages were in English, while only 0.2% were in Hindi, a language spoken by more than half a billion people.

The gaps can reach into daily use. One study of training data for Quechua, a language spoken today by about 8 million people mainly across South America, found that it included texts from the era of Spanish colonization. Developers of most frontier models rely heavily on publicly available internet data, which contains biases and harmful content. Research on AI assistants has also found that when users in India write with these tools, their writing styles shift to mimic Americans and their language can lose cultural nuance. Cost can vary by language as well. AI companies often charge by the token, a unit of text, and the same message can require more tokens in Hindi and Odia than in English. Newer systems have narrowed that gap but not eliminated it, leaving developers building apps in those languages to pay more to process the same content.

Access itself can be disrupted by policy decisions made far away. In June, citing national security concerns, the U.S. government ordered Anthropic to block foreign nationals, including members of its own staff, from accessing its Fable 5 and Mythos 5 models. Anthropic responded by taking both models offline for all users. The restrictions were lifted later that month, but the episode showed how quickly a single government's decision can cut off access to technology used around the world.

That vulnerability has pushed several middle powers to build sovereign AI models of their own. Sarvam in India, Cohere in Canada, Mistral in France and Jais in the United Arab Emirates have developed systems designed to better reflect local needs, contexts and languages. The governments of India and Canada are moving to invest in domestic AI firms, including through equity stakes, to gain more control over the technology and protect national interests. In the UAE, Jais was developed through a partnership involving Inception, a subsidiary, reflecting how states are trying to combine public priorities with private expertise.

The central question is who will have a voice in the systems that are becoming part of everyday public life. For most countries, the answer for now depends on choices made in the United States and China, on the energy, chips and data those two nations control, and on whether independent efforts can build capacity fast enough to matter.

Same event, other desks

Story file →