China Redraws the Geopolitical Balance of AI by Exporting Open Source as a New Technology Diplomacy

Chinese artificial intelligence models now represent 30% of global AI technology usage, compared to just 1.2% at the end of 2024. This meteoric rise reveals an unprecedented geopolitical strategy: Beijing is transforming open source into a diplomatic weapon to redistribute the world’s technological cards. While the United States maintains its closed proprietary models, China opens its algorithms to the entire world and reaps geopolitical influence that Washington had not anticipated.

This new form of technology soft power challenges the foundations of Western leadership in innovation. The performance gap between the best American and Chinese models has narrowed to just 2.7% as of March 2026, according to Stanford’s AI Index. American export controls on chips, far from hindering Chinese laboratories, have paradoxically accelerated their algorithmic innovation and consolidated their global open source strategy.

The Figures Behind a Silent Redistribution of Cards

A study by MIT and Hugging Face reveals that Chinese open source models represent 17.1% of global AI downloads in the year ending August 2025, surpassing the American share of 15.86% for the first time. This historic inversion masks an even more striking reality: on OpenRouter, the world’s largest AI model aggregation platform, Chinese models represented 61% of total token consumption among the ten most-used models in February 2026.

The scope of this transformation exceeds technical metrics. Hugging Face data shows that Alibaba now has the largest number of user-generated model variants, surpassing Google and Meta models combined. This proliferation of derivatives reflects massive adoption by the global developer community, creating a technological ecosystem increasingly dependent on Chinese foundations.

The economic dominance of this strategy becomes measurable. In December 2025, downloads of Alibaba’s Qwen surpassed the combined total of the next eight competitors: Meta, DeepSeek, OpenAI, Mistral, Nvidia, Zhipu.AI, Moonshot, and MiniMax. With more than 700 million downloads for the Qwen family and more than 180,000 derivative models created, Alibaba has become the world’s largest supplier of open source AI systems.

Innovation Under Constraint, an Unexpected Accelerator

American sanctions produced the opposite of the intended effect. Far from paralyzing Chinese innovation, the embargo on Nvidia chips acted as a technological accelerator, forcing Chinese laboratories to develop hyper-efficient software architectures that extract exponentially more intelligence per computational operation than Western models.

The DeepSeek R1 model illustrates this revolutionary efficiency. Trained for approximately 6 million dollars compared to around 100 million for GPT-4, it uses a mixture-of-experts architecture that activates only 37 billion of its 671 billion parameters per inference. This optimization allows Chinese models to operate at approximately one-sixth to one-quarter of the cost of comparable American systems, with an API price of $0.028 per million tokens, roughly 1/180th of the equivalent price of GPT.

Even more significantly, Zhipu’s 744-billion-parameter GLM-5 model was entirely trained on Huawei Ascend chips using the MindSpore framework, proving that frontier-level AI can be synthesized and scaled entirely outside the established Nvidia CUDA ecosystem. This complete technological autonomy redistributes the balance of power in AI geopolitics.

The Global South Adopts the Chinese Alternative

China’s strategy finds its most receptive audience in Global South countries, where Chinese enterprises are well positioned to address digital sovereignty priorities. Singapore’s AI Singapore program, supported by the Singapore government, chose Alibaba’s Qwen rather than Meta’s Llama to build its latest regional model. Malaysia announced that its sovereign AI ecosystem would run on DeepSeek.

This massive adoption reveals a deeper geopolitical fracture. Despite Western criticism, large portions of the Global South embrace Chinese models, viewing them as a path to AI sovereignty. From founders in Nairobi to São Paulo to San Francisco, entrepreneurs are building their services on Chinese foundations.

The reasons for this preference transcend technical considerations. Chinese has become the second most-used prompt language globally, representing nearly 5% of all queries behind English. This percentage significantly exceeds the share of Chinese on the internet, estimated at approximately 1.1%. This overrepresentation testifies to the massive adoption of Chinese models far beyond China.

Qwen supports 119 languages and dialects, reaching deeply into Southeast Asian languages, Arabic dialects, African languages, and regional variants that major Western companies generally ignore. This extensive language coverage strategy deliberately targets emerging markets neglected by Silicon Valley.

Technology Soft Power Against Democratic Leadership

For China, AI is a centralized geopolitical infrastructure, sovereign and aligned with its Belt and Road-style diplomacy. It emphasizes sovereign computing power, data control, and state-directed development. This vision directly opposes the American model, which views AI as an economic engine and pillar of national security, anchored in open innovation, private enterprise, and alliances between democracies.

This strategy responds to a high-level political imperative: during the Politburo study session in April 2025, Xi Jinping called on the country to “vigorously engage in international cooperation on AI” and to “help Global South countries improve their technological capabilities.” These technologies will fundamentally reshape policing, education, health, legal services, governance systems, and regulatory regimes across much of the world, conferring considerable soft and hard power to the countries that develop and distribute them.

The American response remains primarily defensive. Washington has devoted relatively few resources to this problem, favoring protective measures, particularly export controls, to constrain Chinese AI advances. It has accorded only selective attention to foreign promotion, focusing on a handful of wealthy dictatorships seeking to play a major role in cutting-edge AI development, while neglecting a much larger group of Global South nations with growing technological needs.

The Limits of an Attractive but Risky Strategy

This Chinese expansion carries significant shadows. Chinese models bear the imprint of China’s content moderation regime and are trained to avoid outputs that conflict with government policy. In February, Anthropic accused several Chinese laboratories of illicitly extracting Claude’s capabilities through distillation, using fraudulent methods according to leading American companies.

Security vulnerabilities constitute a major issue. The National Institute of Standards and Technology tested DeepSeek’s R1 model and found it complied with 94% of overtly malicious requests using common jailbreaking techniques, while comparable American models complied with only 8%. Google Threat Intelligence identified malware strains that query Qwen models for real-time code generation during active intrusions. AI-assisted cyberattacks have increased by 72% since 2024.

Beyond technical risks, China’s 2017 National Intelligence Law requires companies to “support, assist and cooperate with state intelligence work.” Users sharing contracts, code, and strategic documents with these systems are effectively depositing them in a database accessible to the Chinese government.

Toward a Multipolar AI World

Southeast Asia is already adopting a dual-stack technology approach: American hyperscaler infrastructure overlaid with Chinese open source models. Singapore’s OCBC operates DeepSeek and Qwen alongside Google’s Gemma for different internal functions. Malaysia launched a sovereign AI ecosystem on Huawei GPUs. Indonesia’s Indosat is building telecom AI solutions on DeepSeek.

This geopolitical pragmatism draws the outlines of a fragmented AI future. As the Global South navigates between these competing visions, the result may not be a clean division, but a fractured landscape is inevitable. The coming decade will depend on which model proves most persuasive, most scalable, and most aligned with global aspirations.

The redistribution of the world’s technological cards is underway. Widespread global adoption of Chinese open source models could reshape global patterns of technological access and dependency, and impact AI governance, security, and competition. For the first time since the emergence of the Internet, the United States is losing its monopoly on fundamental digital infrastructure. Chinese open source is not merely distributing code: it is redistributing geopolitical influence in tomorrow’s digital world.

Sources

Sources: