The Dragon's AI Engine: Unpacking China's Global Ambitions and the Rise of Propaganda-Laden AI
The People's Republic of China (PRC) is engaged in a sweeping, state-directed campaign to dominate global artificial intelligence (AI). This ambitious endeavor is fueled by a massive infrastructure expansion, a deliberate strategy of military-civil fusion, and targeted international engagement, all viewed by the Chinese Communist Party (CCP) as fundamental to both economic transformation and military modernization. Recent reports shed light on the astonishing scale of this buildout and its concerning implications, including the potential for AI models to become conduits for CCP narratives.
China's AI Infrastructure Surge: A National Mandate
At the heart of China's AI ambition is its rapidly expanding AI infrastructure, particularly its data centers. The PRC aims to deploy 105 EFLOPS of AI compute by 2025, which represents 35% of its total projected compute capacity of 300 EFLOPS. To achieve this, China has embarked on a national data center construction spree.
Key insights into this infrastructure surge include:
- Massive Scale: As of mid-2024, the PRC had built or announced plans to build more than 250 AI data centers across the country. This is projected to result in over 750 EFLOPS of overall compute capacity, significantly exceeding official targets. Strider identified 207 PRC-owned AI data centers, with 101 announced and 106 already operational.
- Rapid Growth: The number of both announced and operational data centers more than doubled from 2023 to 2024.
- Strategic Geographic Distribution: Driven by the "East Data, West Compute" policy, China is rebalancing its computing capabilities, leveraging the abundant clean energy potential in western regions for "large" and "super-large" data centers, while eastern regions host hubs for more intensive computing needs. While every administrative region except Tibet has at least one AI data center, over half are concentrated in just ten regions.
- Global Reach: The PRC's strategy extends beyond its borders. Policy documents call for building "overseas computing facilities" in Belt and Road Initiative (BRI) countries. Two such AI data centers have already been identified outside the PRC: one in Jakarta, Indonesia, and another in Pasig, Philippines. These overseas facilities could help mitigate internal energy constraints, skirt U.S. hardware export controls, and provide additional compute capacity for People's Liberation Army (PLA) forces operating abroad.
Compliance Hub
Military-Civil Fusion: AI as a Dual-Use Weapon
A critical aspect of China's AI strategy is its military-civil fusion (MCF) policy, which views AI as a path to "all-element, multi-domain, highly efficient new pattern of civil-military integration". The PLA is a direct and central beneficiary of this hardware and software infrastructure.
- Deep Connections: Analysis shows that at least 88 of the 856 stakeholder entities involved in the PRC's AI data center buildout have documented ties to either the PLA, the PRC's defense industrial base, or U.S.-sanctioned organizations. Specifically, 59 have ties to the PLA (including as registered suppliers or U.S.-recognized "Chinese military companies"), 53 have ties to the PRC defense industrial complex, and 29 have relationships with U.S.-restricted entities.
- Military Applications: New, large-scale AI data centers are servicing military applications, with Chinese developers offering AI products for unmanned systems and intelligent command platforms. The Chengdu Intelligent Computing Center, for instance, openly promotes its role in supporting PLA modernization through "smart military" solutions, including UAV development, military database management, and visual data analysis for combat.
- Embodied Intelligence: PRC planners are also focused on developing "embodied intelligence" systems – AI that can physically interact with and adapt to the real world through robots or autonomous vehicles. The PLA is actively exploring how to incorporate embodied intelligence into its operations, seeing it as a way to create "a more flexible and efficient combat system".
- Dual-Use Capabilities: Many customer-facing PRC AI applications have clear potential for surveillance and propaganda. Case studies highlight companies like:
- Xiamen Yuanting Information Technology (DataExa), which developed the Tianji Military Model Platform, an LLM designed for national defense, enabling multi-domain joint combat operations across land, sea, air, space, electromagnetic, and cyber domains.
- Beijing Huaru Technology Ltd., which created XSimVerse, a next-generation military AI large model supporting 25 AI-driven military applications, including intelligent command decision-making and combat simulation.
- Zhipu AI, a spin-off from Tsinghua University, whose ChatGLM series LLMs are linked to the PRC military and defense establishment and was recently placed on the U.S. Entity List due to concerns about its support for China’s military modernization.
- SDIC Intelligence (formerly Xiamen Meiya Pico Information Company), sanctioned by the U.S. for its involvement in developing spyware, which boasts "partnership with overseas companies" in 29 countries despite its ties to the PRC's intelligence apparatus.
Compliance Hub
The Propaganda Problem: AI Models and CCP Narratives
Beyond military applications, a significant concern is the ingrained bias toward CCP viewpoints observed in popular AI models, even those hosted in the U.S.. A report by the American Security Project found that five popular large language models (LLMs) – OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini, DeepSeek’s DeepSeek-R1, and X’s Grok – sometimes returned responses indicative of censorship and bias aligning with the CCP.
- Evidence of Bias: When prompted about the Tiananmen Square massacre (June 4, 1989), most LLMs used passive voice, avoided specifying perpetrators or victims, and described state violence as a "crackdown" or "suppression". When prompted in Chinese, DeepSeek and Copilot used Beijing’s preferred term "The June 4th Incident," while only ChatGPT used "massacre".
- Root Cause: Training Data: This bias stems from the AI models being trained on global information environments that collect, absorb, and internalize CCP propaganda and disinformation, often presenting it with the same credibility as factual information. The training data incorporates Chinese characters used in official CCP documents and reporting, which are "exactly mirrored" in some model responses, particularly DeepSeek and Copilot, demonstrating direct absorption of CCP narratives.
- Propaganda Outlets Using AI: Chinese state-affiliated media, such as Sichuan Daily Network Media Development, a major hub for Chinese propaganda, are actively developing AI-powered platforms for "international communication" to enhance content accuracy and enable "one-button publication to multiple accounts on major overseas social media platforms". This highlights the potential for AI-driven cognitive warfare, which has "serious potential security impacts".
Implications and Recommendations
The PRC's AI infrastructure strategy is not merely a market trend; it is a state-driven, globally networked campaign to gain enduring asymmetric advantage by fusing commercial capacity with geopolitical intent. To safeguard technological leadership and counter these threats, a coherent, whole-of-nation response is vital.
Recommendations for Policymakers include:
- Restricting Security-Relevant AI Infrastructure Developers: Continuously monitor and restrict entities with ties to the PLA and other PRC national security organizations to prevent U.S. technology and know-how from falling into the wrong hands.
- Strengthening International Collaboration: Launch a multilateral effort with Five Eyes allies (U.S., UK, Canada, Australia, New Zealand) and other like-minded nations to track PRC AI infrastructure projects abroad, particularly in BRI countries. This should involve intelligence sharing, technical assistance, and coordinated restrictions on risky AI research partnerships.
- Leveraging Dependence on Foreign Software: Develop targeted export controls and licensing regimes for critical software tools used in AI model training, data center operations, and systems integration that are embedded in PRC AI infrastructure. Coordinate with allies to restrict updates and technical support for software used in military-linked or dual-use facilities.
Compliance Hub
Recommendations for Industry Leaders and Investors include:
- Monitoring PRC Activity: Track PRC statecraft tactics in near real-time to identify intersections with people and technology, enabling proactive risk mitigation.
- Eliminating Vectors of Technology Transfer: Avoid relationships with both direct PRC competitors and economic statecraft entities that support them.
- Enhancing Awareness: Proactively engage with top talent to raise awareness about PRC tactics targeting AI experts globally.
- Understanding the Competitive Landscape: Identify not only direct PRC competitors but also the broader network of government research institutes, universities, industry associations, and technology parks that support China's AI strategy. Assess risk by mapping ties to the PRC military and state, and mitigate against talent recruitment and IP loss.
By adopting these practices, policymakers and industry leaders can fortify U.S. and allied AI industries, protect critical technologies, and secure long-term strategic and economic competitiveness against China's formidable AI ambitions.
