China AI Power Actors

Quick Assessment

Key Links

Overview

China's artificial intelligence ecosystem is not governed by a single agency or institution but by a layered network of Party-state organs, central ministries, military bodies, state-affiliated research institutes, private technology companies, and local government innovation zones. Together these actors implement the Chinese Communist Party's strategic goal of achieving global AI leadership by 2030, as originally articulated in the 2017 New Generation Artificial Intelligence Development Plan issued under Xi Jinping. The system combines top-down ideological direction with market-oriented private-sector competition, producing a hybrid structure that differs substantially from both the US model of predominantly private-led development and the EU model of regulatory-first governance.

The ecosystem's distinguishing feature is the degree to which state priorities — national security, social stability, industrial competitiveness, and technological self-reliance — are embedded into the objectives of nominally private firms. Under China's 2017 National Intelligence Law, companies are required to cooperate with state intelligence activities, and leading firms like Baidu, Alibaba, Tencent, and Huawei have been formally designated as national AI champions tasked with developing platform-level infrastructure in their respective domains. This state-capital nexus enables rapid coordination between government priorities and private R&D capacity, but also subjects companies to political risk and regulatory uncertainty that has periodically undermined investor confidence.

This article focuses on the who of Chinese AI governance — the specific actors and their roles — and complements China AI Regulatory FrameworkPolicyChina AI Regulatory FrameworkComprehensive analysis of China's AI regulatory framework covering 5+ major regulations affecting 50,000+ companies, with enforcement focusing on content control and social stability rather than ca...Quality: 57/100, which covers the regulatory instruments those actors produce and enforce. Understanding this actor landscape is increasingly important for AI safety researchers, policymakers, and anyone seeking to understand how AI development is likely to unfold in the world's most populous country and second-largest economy.

History and Background

China's structured approach to AI as a national strategic priority dates to the mid-2010s. The 2015 "Internet Plus" plan and the 2016 "Three-Year Action Plan for AI" laid early groundwork, but the pivotal moment came with the 2017 New Generation AI Development Plan (新一代人工智能发展规划), which set the explicit goal of making China the world's premier AI innovation center by 2030, with intermediate milestones in 2020 and 2025.

In 2018, the Ministry of Science and Technology (MOST) designated the first wave of "national AI champions," naming Baidu (autonomous driving), Alibaba (smart cities), Tencent (medical imaging), iFlytek (speech), and SenseTime (computer vision / facial recognition) as sector leads. This designation expanded over time to a list of 15 companies tasked with sector-specific platform development. The intent was to concentrate resources, create replicable platforms, and diffuse AI capabilities into the broader economy.

The period from 2022 to early 2025 is characterized by researchers as a "Catch-Up Era" in which China loosened some AI restrictions to spur growth in the wake of US chip export controls imposed in 2022. The release of DeepSeek-R1 in January 2025 marked a turning point — placing a Chinese open-weight model at the global frontier and catalyzing what analysts at the Carnegie Endowment describe as a "Crossroads Era," in which the CCP has expressed renewed confidence in China's AI trajectory and begun accelerating AI deployment across critical infrastructure while also considering tighter governance of AI outputs and safety.

Institutional development on AI safety has been more recent. The Artificial Intelligence Industry Alliance (CAICT) released voluntary safety commitments in December 2024 signed by 17 companies including DeepSeek, Alibaba, and Tencent. In February 2025, the China AI Safety and Development Alliance (CnAISDA) was launched at the Paris AI Action Summit, representing the first dedicated coalition of Chinese institutions focused on catastrophic and long-term AI risks.

Key Activities and Actor Map

1. Party-State Apex

The Chinese Communist Party Politburo Standing Committee sets the overarching direction for China's AI strategy. Xi Jinping has presided over high-level policy speeches framing AI as essential to national rejuvenation, economic modernization, and military modernance. The Politburo held its first dedicated AI development and safety meeting since 2018 in a recent session that signaled growing attention to both the opportunities and governance challenges of advanced AI.

The State Council operationalizes Politburo directives through industrial policy documents, funding allocations, and inter-ministerial coordination. The 2017 AI plan, for example, was a State Council document that cascaded into ministry-level implementation programs.

2. Central Agencies

Four central agencies share primary jurisdiction over AI governance, often with overlapping and sometimes competing mandates:

The Cyberspace Administration of China (CAC)OrganizationCyberspace Administration of China (CAC)China's primary internet and AI regulator. Responsible for implementing China's generative AI regulations, algorithm registration requirements, and deepfake rules. has emerged as the most active regulatory actor, having introduced regulations on recommendation algorithms (2021), deep synthesis / deepfakes (2022), and generative AI services (2023). It requires platforms like Douyin and Bilibili to label AI-generated content and mandates security assessments before public deployment of generative models.

3. Military Actors

The People's Liberation Army (PLA), particularly through bodies that conduct military-civil fusion research, is a significant AI actor. China's Military-Civil Fusion (MCF) strategy — institutionalized under Xi Jinping — directs civilian technology firms and universities to make their AI research available for military applications without formal nationalization. The National University of Defense Technology (NUDT) conducts AI-oriented research in areas including autonomous systems, AI-accelerated simulation, and intelligent command systems.

China's military AI activities are of particular concern to US policymakers. The US China AI Power Report Act (H.R. 6275), introduced in November 2025 by Rep. James Moylan (R-GU), would mandate annual Commerce Department assessments of PRC AI capabilities specifically covering military AI models and AI systems used for surveillance of Uyghurs and other ethnic minorities.

4. State-Affiliated Research Labs

Several state-affiliated institutions function as the primary basic-research backbone of China's AI ecosystem:

• Beijing Academy of Artificial Intelligence (BAAI / 智源研究院): A flagship institute supported by the Beijing municipal government and MOST, BAAI has produced major open-weight models and serves as a convening hub for Chinese AI researchers. It focuses on large-scale pretraining, embodied AI, and AI safety research.
• Institute of Automation, Chinese Academy of Sciences (CASIA): One of China's oldest and largest AI research institutes, CASIA conducts research on computer vision, pattern recognition, and brain-inspired computing, with deep ties to state security applications.
• Shanghai Artificial Intelligence Laboratory (Shanghai AI Lab): A newer institution with significant municipal and central government backing, Shanghai AI Lab has produced open-source models and participates in the CnAISDA safety coalition. It has focused on multimodal AI and scientific AI applications.

5. Private Technology Giants

China's leading technology companies function as the primary deployment layer for AI, with each assigned a specific national platform domain:

• Baidu: Designated national champion for autonomous driving and AI search. Its ERNIE (文心) large language model family is among China's most widely deployed commercial LLMs, integrated into Baidu Search (which reportedly attracted 10 million users in one hour following a DeepSeek integration). Baidu has the longest track record of large-scale AI R&D among Chinese internet firms.
• Alibaba: Designated national champion for smart cities and cloud AI infrastructure. Its Qwen model family has become the world's largest open-source AI system by download count, surpassing 700 million downloads. Singapore's national AI program selected Qwen as its foundation model — a notable instance of Chinese open-source AI gaining state-level adoption abroad.
• Tencent: Designated champion for medical imaging AI; its Hunyuan LLM debuted in 2023. Tencent operates WeChat, giving it unmatched data on Chinese social communications, which informs its AI product development.
• ByteDance: Though not in the original 2018 champion list, ByteDance has emerged as a major AI actor through its Doubao large model and assistant product, which has achieved ubiquity in China comparable to ChatGPT's position in the US — a dynamic that Western observers frequently overlook when assessing China's AI adoption.
• iFlytek: Designated national champion for speech recognition and natural language processing, with deep ties to state security and education applications.
• SenseTime and Megvii (Face++): Leading computer vision companies specializing in facial recognition; both have been subject to US Entity List designations citing their role in surveillance of Uyghurs and other ethnic minorities.

6. Emerging Model Developers ("AI Tigers")

A second generation of AI-focused companies has risen to prominence since 2023, often described collectively as China's "AI Tigers":

DeepSeek's trajectory is the most dramatic: its R1 model's January 2025 release prompted high-level CCP meetings with AI pioneers and accelerated government commitments to AI deployment across critical infrastructure. According to Carnegie Endowment analysis, this success shifted the policy environment from cautious encouragement to active promotion, though it also raised the likelihood of tighter content controls.

7. Chip Ecosystem

China's semiconductor actors are central to understanding the ecosystem's constraints and strategic responses:

• Huawei (Ascend chips): Huawei's Ascend AI accelerator series represents China's most advanced domestically produced alternative to NVIDIA GPUs. Following US export controls in 2022 that restricted access to advanced NVIDIA chips, Huawei's Ascend 910B and subsequent variants have become the primary high-performance training option available to Chinese developers. Huawei has also integrated DeepSeek for African markets, signaling potential AI-focused Belt and Road deployments.
• SMIC (Semiconductor Manufacturing International Corporation): China's leading contract chipmaker, constrained by US restrictions from accessing advanced EUV lithography equipment, which limits its ability to manufacture at the most advanced process nodes. SMIC nonetheless continues to supply mature-node chips for a wide range of applications.
• Cambricon: A specialized AI chip designer that produces inference accelerators used in cloud and edge deployments; it has government backing and university origins (Chinese Academy of Sciences).

The chip ecosystem represents the most significant structural constraint on China's AI development. US export controls have denied access to NVIDIA's most advanced training hardware, forcing Chinese developers to work with older NVIDIA chips stockpiled before controls took effect, Huawei Ascend alternatives, and — as DeepSeek demonstrated — algorithmic efficiency optimizations that reduce compute requirements.

8. Local Government Innovation Zones

Sub-national governments play an underappreciated role in China's AI ecosystem by providing land, subsidies, preferential tax treatment, and regulatory sandboxes:

• Beijing: Home to BAAI, NUDT research affiliates, Baidu, ByteDance, and most of China's leading AI startups; the Zhongguancun Science Park functions as China's primary AI innovation hub.
• Shanghai: Hosts Shanghai AI Lab, major Alibaba and Tencent R&D centers, and the WAIC (World Artificial Intelligence Conference), an annual government-organized showcase of AI progress.
• Shenzhen: Headquarters of Huawei and Tencent; a manufacturing hub for AI hardware and consumer devices integrating AI, including DJI drones.
• Hangzhou: Alibaba's home city; a center for cloud AI, e-commerce AI, and smart city infrastructure.

Regional specialization is increasingly pronounced: Beijing concentrates on foundational model research and policy-adjacent work, Shanghai on international-facing AI governance and multimodal systems, Shenzhen on hardware and industrial AI, and Hangzhou on cloud and enterprise applications.

AI Safety Actors

China's AI safety ecosystem is less mature than its US counterpart but has developed rapidly since 2024. Key actors include:

• CnAISDA (China AI Safety and Development Alliance): Launched February 2025 at the Paris AI Action Summit; brings together experts and institutions including Shanghai AI Lab for testing and evaluation. It focuses on catastrophic and long-term AI risks and emphasizes international cooperation.
• Artificial Intelligence Industry Alliance (CAICT): Released voluntary safety commitments in December 2024, signed by 17 companies. These commitments mirror the structure of global Frontier AI Safety Commitments.
• Academic voices: Scholars such as Andrew Yao (姚期智) have voiced concerns on AI alignment, cited alongside Geoffrey Hinton as representing shared US-China worries. IR scholar Lu at Tongji University has argued for categorizing AI existential risks as national security concerns and proposed a coordinating institution modeled on the US AI Safety Institute.
• Framework 2.0: China's AI standards roadmap, which explicitly addresses loss of control scenarios via mechanisms described as "circuit breakers and one-click control" for extreme situations. The framework acknowledges risks including AI weapons of mass destruction and uncontrollable AI behavior.

China does not yet have a formal AI Safety Institute and is not part of the US-UK international AI safety network, though the CnAISDA represents a move toward institutionalizing safety concerns.

Criticism and Concerns

Several significant criticisms attach to China's AI actor ecosystem, spanning authoritarian use cases, governance gaps, and geopolitical risks.

Surveillance and repression: SenseTime, Megvii, and iFlytek have all been connected to AI systems used in the surveillance and monitoring of Uyghurs and other ethnic minorities in Xinjiang. US legislation including H.R. 6275 specifically mandates annual reporting on AI models used for this purpose. Critics argue that the state-capital nexus structurally incentivizes AI companies to develop capabilities for social control that would face regulatory barriers in democratic countries.

Governance-innovation tension: Analysts at the Carnegie Endowment note that China's AI policy oscillates between growth-oriented permissiveness and ideological control, and that heavy-handed content controls can stifle economic integration of AI. Erratic tech policy has periodically deterred investors and imposed compliance burdens — content moderation requirements, data localization rules, security assessments — that disadvantage Chinese firms relative to less-regulated global competitors.

Diffusion weaknesses: According to Jeffrey Ding's analysis, China prioritizes elite AI innovation over broad STEM education and workforce training, which may limit economy-wide AI diffusion relative to the United States. The gap between frontier model capability and broad economic deployment remains a structural challenge.

Open-source security risks: The rapid global proliferation of Chinese open-source models has introduced security concerns. Google Threat Intelligence has identified malware strains querying Qwen models for real-time code generation during active intrusions. AI-assisted cyberattacks reportedly increased 72% since 2024, and Chinese open-source models' relative lack of strong terms of service or safety filtering has been cited as a risk factor by security researchers.

Geopolitical dominance concerns: EA Forum and LessWrong community discussions reflect concern that if CCP-aligned AI achieved global dominance, it could embed censorship, surveillance, and propaganda capabilities into global AI infrastructure in ways that conflict with democratic values and AI safety norms. These discussions generally view a US lead in transformative AI as preferable from an AI safety standpoint, while acknowledging this is contested.

Key Uncertainties

• Whether US chip export controls will effectively constrain China's frontier model training capacity, or whether algorithmic efficiency improvements and domestic chip production will compensate
• Whether the CnAISDA and related safety institutions will develop technical depth comparable to US/UK AI safety bodies, or remain primarily rhetorical
• How the tension between CCP ideological control and economic imperatives will resolve as AI becomes more central to economic growth
• Whether China's open-weight model dominance in the Global South will translate into durable geopolitical influence via embedded AI standards
• The actual extent of military AI capability, which is not publicly verifiable