CNAS data shows sovereign AI initiatives spread to 67 countries and the EU
America is facing a critical juncture in the global artificial intelligence race as foreign powers deploy open-source models to capture international markets. According to data analyzed by the Center for New American Security (CNAS) and reported by thecipherbrief.com, sovereign AI initiatives have spread to 67 countries and the European Union by mid-2026, creating a strategic challenge for Washington’s technological leadership.
Xi Jinping Pitches Open Source in Shanghai
Speaking in Shanghai on July 17, Xi Jinping outlined a vision for artificial intelligence centered on open source, openness, and sharing, while cautioning against expansive national security concepts. Xi announced that China will train 5,000 individuals from developing countries over the next five years. Beijing is also building AI application cooperation centers with major regional blocs, including ASEAN, the Arab League, the African Union, CELAC, the Shanghai Cooperation Organization, and BRICS, alongside the expansion of its MAZU meteorological warning system to 30 countries. The World Artificial Intelligence Cooperation Organization now operates out of Shanghai with 37 member states, providing a formal institutional framework for these efforts.
This diplomatic push pairs inexpensive inference and capable open-weight models with public goods, institutional backing, and the vocabulary of multilateralism. For governments seeking local control over digital infrastructure rather than reliance on a foreign cloud application programming interface (API), a locally operated model represents an appealing alternative.
Global Adoption Shifts Toward Chinese Foundations
The practical results of Beijing’s strategy are already evident in international infrastructure projects. Most sovereign AI model projects that disclose their foundations choose to adapt an existing open-weight model rather than train a new one from scratch, and an increasing share of those foundations are Chinese. Spain’s Quasar 438B, marketed as Europe’s strongest reasoning model, relies on Z.ai’s GLM-5.2 architecture. Japan’s Rakuten AI 3.0 utilizes the DeepSeek-V3 framework, while AI Singapore constructed its regional model on Alibaba’s Qwen. In the Middle East, Saudi Arabia’s HUMAIN commissioned its Arabic language model from MiniMax.
Usage metrics on developer platforms reflect this shift. Chinese open-weight models surpassed the American share of global downloads for the first time in 2025. On OpenRouter, a marketplace routing developer traffic to various models, Chinese model usage climbed from under 15 percent to more than half within a single year, driven heavily by businesses across the Global South.
China Links AI to Political Security and Military Transformation
Writing in the Cyberspace Administration’s magazine in September, China’s Minister of State Security Chen Yixin described AI as a primary battleground for great-power competition, linking technological independence and control over international governance to political security, cyber capabilities, and military transformation.
The institutional mechanisms deployed by Beijing serve these dual objectives of technical integration and strategic influence. An Egypt-specific variant of the MAZU meteorological warning system is currently under development to handle aviation weather and sandstorm tracking. Simultaneously, Huawei is engaging officials in Cairo regarding its Ascend chips, while Malaysia evaluates the same hardware stack.
Competing Across the Entire Technology Stack
Maintaining leadership requires the United States to compete across hardware, models, software, and services rather than retreating exclusively to frontier models and advanced chips. If adversary nations supply the open models, affordable inference, and industrial applications where actual adoption occurs, technological dominance at the frontier becomes irrelevant to daily global usage.
Openness alone does not guarantee preference. American models must outperform competitors on total cost, local-language capabilities, reliable supply chains, and performance, while allowing domestic firms in partner nations to build viable commercial enterprises. Regulatory stability is paramount. When U.S. export controls forced Anthropic to suspend access to its newest models from June 12 to June 30, international governments observed that even close allies could experience sudden interruptions in American systems based on policy decisions made in Washington.
The Evolving Regulatory Landscape
Beijing’s open-source posture may face internal constraints as domestic regulators reassess the risks of capability diffusion. Reuters reported in July that Chinese authorities were weighing restrictions on their own frontier models. The TC260 AI Safety Governance Framework 3.0, released on September 14 under the guidance of the Cyberspace Administration of China, addresses autonomous cyberattacks, loss of control, and removable safeguards, suggesting that China may adopt a more selective approach to open releases.
To secure a lasting advantage, Washington can build on existing mechanisms such as the July 2025 AI Action Plan and Executive Order 14320, which supports exporting the American AI stack through coordinated technology packages, diplomacy, and financing. Targeted investments in undershot market segments—such as sustained open releases, smaller-language models, and affordable local deployment—will determine whether partner nations choose American infrastructure to power their digital futures.