World News

The Bipolar AI Race: Nations Seek Alternatives to U.S.-China Dominance

As the U.S. and China consolidate power over artificial intelligence, other countries are investing in sovereign models and coalitions to ensure technological independence.

The global race for artificial intelligence is increasingly defined by a bipolar concentration of power, with the United States and China controlling the vast majority of the infrastructure, research, and governance of the technology. This duopoly has left the rest of the world struggling to gain a foothold in how AI systems are built, tested, and regulated.

Data from Epoch AI, a research organization, indicates that between 2022 and June 2026, the United States secured approximately 72% of the world’s new AI computing capacity, while China received 10%, leaving only 18% for the rest of the world. This concentration extends to physical and intellectual resources as well. The International Energy Agency estimates that in 2024, the U.S. and China accounted for 70% of global electricity consumption by data centers. Furthermore, a 2025 Carnegie Endowment study of more than 10,000 researchers who published at the NeurIPS AI conference found that three-quarters were affiliated with institutions in the U.S. or China.

Beyond infrastructure, the global impact of these technologies is often limited by a lack of linguistic and cultural representation. A recent snapshot of Common Crawl, a web archive used for training language models, showed that 43% of its pages were in English, while Hindi—spoken by more than half a billion people—accounted for a mere 0.2%. Research has also highlighted that frontier models often reflect the biases of their training data, potentially leading to the homogenization of language and cultural nuance in countries like India.

The vulnerability of other nations to policy decisions made in Washington or Beijing was starkly illustrated in June, when the U.S. government, citing national security concerns, ordered Anthropic to block foreign nationals from accessing certain models. Anthropic complied by taking the systems offline, though the restrictions were later lifted. The incident accelerated efforts by middle powers to develop “sovereign” models, including India’s Sarvam, Canada’s Cohere, France’s Mistral, and the UAE’s Jais.

While these models aim to reflect local contexts, their effectiveness is limited by the fact that many are built by adapting open-weight models from the U.S. or China, potentially inheriting existing biases. Additionally, efforts to improve these models through human feedback often lack sufficient representation from the Global South. Even Meta’s Community Alignment project, which sought to include different cultural preferences, drew participants from only two countries and excluded hundreds of languages spoken in India.

In response to this technological divide, some nations are forming coalitions to strengthen their collective bargaining power. The Global Partnership on Artificial Intelligence (GPAI), which includes 46 countries and the European Union, is one such platform. In September, during the 81st session of the United Nations General Assembly, Finland proposed measures to manage AI development and establish transparent safety protocols. However, disagreement persists: while 22 countries signed the proposal, others, such as Nigeria and India, held back, with Nigerian officials citing an urgent need for productivity gains over regulation.

As the U.S. and China build competing coalitions—Beijing through the World Artificial Intelligence Cooperation Organization and Washington through the Pax Silica initiative—other countries face pressure to choose sides. Experts suggest that rather than relying solely on frontier models, middle powers should pool resources for shared regional infrastructure, invest in smaller, specialized models, and collaborate on robust, localized testing and evaluation standards.

Sophia Mitchell

Sophia Mitchell reports on international affairs, global developments, and major events across regions. She brings readers reliable updates and useful context on the people, decisions, and challenges shaping an interconnected world.
Back to top button