The Performance Gap in China's PLA Military AI Kill Chain and Its Implications for U.S.-China Military Competition

Executive Summary

A paper by researchers from the Shandong Corps of China's People's Armed Police (PAP) acknowledged that the PLA's use of military AI remains confined to auxiliary roles like Q&A and content generation, with offensive combat applications still in an exploratory phase. This gap is analyzed as a structural problem stemming not from a lack of technology, but from political caution surrounding Party-military relations and the delegation of command authority. The PLA is shifting its focus from individual weapon performance to building "systems warfare" capabilities that integrate intelligence, AI, communications, and logistics. Consequently, a path of gradual integration and asymmetric catch-up is considered most likely over the next three to five years. Given the rapid progress of Chinese commercial AI firms like Zhipu AI, the current gap is unlikely to become permanent. For South Korea, which lacks direct leverage on this issue, a realistic response involves continuously monitoring changes in the military balance in the Taiwan Strait and Western Pacific and tracking the organizational restructuring of the PLA's Strategic Support Force (SSF), while simultaneously advancing its own roadmap for developing a Korean-style AI command and control system.

I. Analysis of the Current Situation

The Performance Gap in China's PLA Military AI Kill Chain: Status and Implications

1. Background and Developments

The starting point for this analysis is a paper published in July by researchers from the Shandong Corps of the Chinese People's Armed Police (PAP) and the PAP Engineering University in the academic journal *National Defense Science & Technology Industry* [1]. Citing U.S. military data extensively, the paper self-assessed that the kill chain of China's military AI models lags behind Western counterparts [1]. The researchers stated that the PLA's use of AI remains concentrated in "auxiliary areas such as intelligent Q&A and content generation," with its application in offensive combat still in an exploratory phase [1].

This self-assessment can be read as a sober appraisal of the results of the military-civil fusion and AI-first strategies pursued by the Xi Jinping leadership since 2017. According to a Brookings Institution analysis, Beijing has declared its plan to become a world-leading AI power by 2030, and the PLA has also formalized an innovation-driven military modernization strategy [2]. The main implementing body is the Strategic Support Force (SSF), established in 2015 [2]. The SSF was designed to integrally manage space, cyber, electronic, and information warfare capabilities, and efforts to incorporate AI into these domains have been centered around this force [2]. It is also noteworthy that this paper was authored by researchers affiliated with the PAP. The fact that a research institution from a paramilitary organization, rather than from the SSF's regular military line, published an article on the limitations of China's military AI in an open academic journal suggests that a sober self-diagnosis of performance is underway within the country's military research community.

2. Current Situation

The gap identified in the paper extends beyond mere performance metrics. The PLA continues to confine its use of AI to support functions such as information processing and document generation [1]. In contrast, the researchers judge that Western models have advanced to integrating AI across the entire kill chain, from target identification to strike [1]. This connects to the concept of "systems warfare" that the PLA has emphasized. Chen Chia-cheng, a researcher at Taiwan's Institute for National Defense and Security Research (INDSR), has analyzed that the PLA is shifting its focus from the superiority of individual weapon systems to the ability to integrate intelligence, AI, communications, weapons, logistics, and infrastructure into a single combat system [5]. According to his assessment, "The PLA is not concerned with whose weapons are stronger, but with who can connect different weapons to create greater comprehensive combat power" [5]. The war in Ukraine and events involving Iran, which have demonstrated the power of integrated operations and the expansion of targeting to include civilian infrastructure, are also seen as factors accelerating this shift [5].

However, caution still prevails in applications approaching real-world combat. A study simulating a maritime dispute in the South China Sea found that under a rule-based command system, the probability of the China Coast Guard using force was only 0.3% [8]. While this figure is low, it was also noted that from Beijing's perspective, even this level of probability could be considered excessive in sensitive waters [8]. The study also confirmed that the AI was designed not to simply classify targets as 'hostile' or 'harmless,' but to continuously calculate the probability of various intentions—ranging from fishing activities to organized threats—and adjust the level of response accordingly [8]. This suggests that research within the PLA system prioritizes risk-management-oriented AI applications over offensive automation.

Meanwhile, a contrasting trend is discernible in the private sector. Chinese companies such as Alibaba, MiniMax, and ByteDance are leading benchmark rankings in video generation AI, creating areas where they surpass their Western counterparts [14]. There are also user reports that Zhipu AI's GLM-5.2 model performs on par with Anthropic's Claude in real-world evaluations [11]. This indicates that the rapid progress in the private generative AI sector and the lag in the military AI kill chain are proceeding at different paces, suggesting a potential structural bottleneck in the technology transfer channels between the civilian and military sectors within China's AI ecosystem.

3. Key Actors and Interests

The PLA and the Strategic Support Force (SSF)are prioritizing the application of AI to information warfare, command and control, and logistical support, while maintaining a cautious stance on its offensive use at the level of autonomous engagement [2][1]. This can be seen as the result of both technical limitations and an institutional conservatism that seeks to maintain human intervention in the chain of command.

Researchers from the Shandong Corps of the People's Armed Police (PAP) and the PAP Engineering Universityare the actors who publicly pointed out the limitations of their country's military AI in the aforementioned paper [1]. The fact that a research institution affiliated with a paramilitary organization diagnosed its own country's gap by citing U.S. military data demonstrates a research practice that prioritizes internal assessment over external propaganda.

Taiwan's Institute for National Defense and Security Research (INDSR)warns that the PLA's shift to systems warfare necessitates a redesign of Taiwan's entire defense posture [5]. Their core argument is that Taiwan must move beyond preparations centered on individual fighter jets, ships, and missile defense to counter the threat of integrated targeting [5].

The U.S. Policy and Research Community(including Brookings, CFR, and Foreign Affairs) treats the PLA's AI integration efforts as a key variable in the U.S.-China military-technological competition, while simultaneously exhibiting a dual stance by also warning of the risks posed by their own military's over-reliance on AI [2][9][13].

China's Cybersecurity Regulatory Authoritiesthrough channels linked to the Ministry of State Security, are categorizing AI risks into issues of algorithmic opacity, model control risks, threats from AI agents, misuse, and technological hegemony, and are developing a domestic control framework separate from military applications [12]. This can be interpreted as a national, risk-management-first approach that aligns with the PLA's cautious attitude toward AI adoption.

4. Key Issues

The first issue is the nature of the gap. What the paper points to is not the inferiority of the algorithms themselves, but rather a lag in the transition from support functions to offensive combat integration [1]. This indicates that the focus of U.S.-China AI competition is shifting from a race for model performance benchmarks to a competition over the capability for real-world integration within military organizations.

The second issue is the asymmetry in civil-military technology transfer. Unlike Chinese private AI companies, which are achieving results on par with or surpassing the West in specific areas like video generation [14][11], the PLA's kill chain integration falls short [1]. Whether this discrepancy is due to an organizational bottleneck within the SSF or because the reliability and safety standards required for combat AI are much higher than for commercial use is a point that needs to be tracked in the future.

The third issue is the implication for the regional deterrence posture, particularly in the Taiwan Strait. The PLA's shift to systems warfare is already underway [5], and there is a coexisting assessment that even if the pace of AI integration into the kill chain is slow, its integrated operational capabilities themselves are continuously being enhanced. From the perspective of a middle power, it is necessary to view the "lag" in the PLA's military AI not as a simple signal of a reduced security threat, but as one facet of the larger trend toward enhancing its integrated command system. Regional countries, including South Korea, need to continuously monitor the trend in the U.S.-China military AI gap in conjunction with the effectiveness of U.S. technology export controls on China and the credibility of extended deterrence in the region.

II. In-Depth Analysis

The Performance Gap in China's PLA Military AI Kill Chain: Root Causes and Structural Context

1. Analysis of Root Causes

The fundamental reason why the PLA's military AI remains confined to support functions is less a matter of technology itself and more an issue of organization and institutions. The paper by the PAP Shandong Corps researchers states, "Military AI applications are mainly used in auxiliary areas such as intelligent Q&A and content generation" [1]. The implication of this sentence is clear: the working-level organizations handling AI within the PLA are optimized for back-office tasks like data processing and document creation. Integrating AI into the combat command system is a separate problem. For AI to be involved in target identification, application of rules of engagement, and real-time fire control, a politically sensitive decision must first be made on how to distribute discretionary authority between commanders and algorithms. The finding from the South China Sea simulation study—that the probability of the China Coast Guard using force under a rule-based command system was 0.3%—illustrates this caution well [8]. This is followed by the observation that from Beijing's perspective, even this low probability could be judged excessive in sensitive waters [8]. The fact that the AI was designed to calculate probabilities of intent, from fishing activities to organized threats, rather than simply dichotomizing targets as 'hostile' or 'harmless,' can also be read as an approach to reduce political risk [8].

Underlying this is a fundamental constraint of the Party-military relationship. The PLA is not a national army but the army of the Chinese Communist Party. The issue of delegating command authority to AI is directly linked to the control of the Party's Central Military Commission. The structure is such that the leadership must be wary of the possibility that malfunctions or unpredictable outputs could lead to political repercussions. The recent official mention by the Cyberspace Administration of China of AI risks—including algorithmic opacity, model control risks, emerging threats from AI agents, misuse, and technological hegemony—is an extension of this caution [12]. Wang Lihong, deputy director of the Cybersecurity Coordination Bureau, pointed out that "the complex and opaque nature of AI algorithms like deep learning makes it difficult to identify and correct flaws" [12]. While this is a risk perception in the civilian domain, this risk-averse tendency is bound to operate much more strongly in the military sphere.

2. Structural Context

Security Structure: SSF-Centered Integration Efforts and Their Limitations

In 2015, the Xi Jinping leadership established the Strategic Support Force (SSF) to bring space, cyber, electronic warfare, and information warfare capabilities under a single command structure [2]. The Brookings Institution assessed this move, stating, "Not only has Beijing announced an ambitious plan to make China the world's leading AI power by 2030, but the CCP has also unveiled an aggressive innovation-driven strategy for the PLA" [2]. However, creating an organization and having that organization actually integrate disparate technologies, personnel, and cultures are two different things. It has been consistently pointed out that the SSF, as a new organization formed by transferring personnel and assets from various services, does not have a homogeneous internal command culture. The fact that the recent paper came from a research institute under the PAP, not from the SSF's regular channels, is also suggestive. It can be interpreted as evidence that military AI integration is proceeding sporadically across various services and PAP organizations, beyond the single axis of the SSF, and that the variation among them is significant.

Strategic Structure: The Gap Between the "Systems Warfare" Concept and Execution Capability

The concept the PLA is aiming for is already quite advanced. Chen Chia-cheng of Taiwan's INDSR analyzed that the PLA is moving toward "integrating intelligence, AI, communications, weapons, logistics, and infrastructure into a single comprehensive combat system" rather than focusing on individual weapon systems [5]. He diagnosed that "the PLA is not concerned with whose weapons are stronger, but with who can connect different weapons to create greater comprehensive combat power" [5]. This is the result of the PLA learning from the power of integrated operations and the trend of target expansion to include civilian infrastructure, as seen in the war in Ukraine and events involving Iran [5]. The problem is the gap between this conceptual orientation and the actual level of its algorithms, data, and command and control infrastructure. Compared to Western models that integrate AI across the entire kill chain, the PLA can be seen as having an advanced concept but lacking the real-world datasets and validated algorithms to support it.

Economic and Industrial Structure: The Asymmetry Between Civilian AI Capabilities and Military Application

China's private AI industry is already showing signs of catching up to or surpassing the United States in certain areas. Companies like Alibaba, MiniMax, and ByteDance are taking the lead in video generation benchmarks [14], and Zhipu AI's GLM-5.2 is reportedly performing on par with Anthropic's Claude according to users [11]. This surge in the private sector is driven by improved training efficiency and a low-cost structure [11], reaching a point where it is pressuring Silicon Valley to lower prices [11]. However, this civilian capability does not automatically transfer to the military domain. The capabilities of large language models, which excel at content generation or Q&A, have fundamentally different requirements from the robustness and verifiability needed for real-time target identification and fire control. Although a system for "civilian participation in military application" (mincanjunyong) exists, it is possible that, unlike in the West, the PLA's internal procurement, certification, and security review processes act as a bottleneck, delaying the rapid military application of cutting-edge civilian models.

3. Historical Precedents and Comparative Cases

The pattern of new military technologies being adopted first for support functions, with integration into core combat functions coming later, is not a new phenomenon. During the Cold War introduction of computer-based command and control (C2) systems, the U.S. military first automated logistics and information processing, taking a much more cautious approach to involving computer models in actual combat decision-making. In the discussions surrounding the Soviet "Perimeter" system, which sought to combine nuclear weapons with automated command and control, the gap between technical feasibility and political acceptability delayed its operational deployment. Existing analysis on the AI-nuclear nexus points out that the military use of AI creates a "multiplier effect on operational speed" [7], but emphasizes that this multiplier effect does not immediately lead to the delegation of command authority. It suggests that a significant period of verification and a change in organizational culture are necessary before commanders can trust and cede discretion to an algorithm's judgment.

An analysis in *Foreign Affairs* raises concerns that even within the U.S. military, the adoption of AI may not necessarily lead directly to enhanced combat power. The risk is that in the process of changing personnel structures and decision-making systems, AI could actually hollow out the organization's judgment [13]. This suggests that the problem is not unique to the PLA but is a general dilemma inherent in military AI integration itself. However, in the PLA's case, the additional variable of the Party's priority on control is superimposed, making it likely that the resolution of this dilemma will be slower than in the West.

4. Key Variables Shaping Future Developments

The first variable is the pace of internal organizational reform within the PLA, including the SSF. For the concept of an integrated combat system to be realized at the level pointed out by Chen Chia-cheng [5], the standardization of data sharing systems and command and control infrastructure among the services must come first. How quickly this work proceeds will determine the actual progress of AI integration into the kill chain.

The second variable is the speed of military application of civilian AI models and the associated verification system. The key is how quickly the technological advances from the private sector, such as Zhipu AI and Alibaba [11][14], are absorbed through the military procurement system, and whether security reviews and reliability verification act as a bottleneck in that process.

The third variable is the dynamic of the U.S.-China military AI competition. As a CFR analysis points out, the U.S.-China AI race is already unfolding so dynamically that it has been described as having "entered a new model" [9]. If the U.S. military accelerates the integration of AI across its kill chain, the PLA could face pressure to increase its own pace of real-world integration, despite its political caution. How quickly this competitive interaction erodes the PLA's organizational cautiousness will be a variable that determines whether the gap narrows in the future.

The fourth variable is whether actual tensions arise in the region, including in the Taiwan Strait. The pressure of a real-world crisis can force a pace of integration not seen during peacetime research and development. A path similar to how the war in Ukraine reshaped the Russian military's integrated operations capabilities in the midst of combat cannot be ruled out for the PLA in the event of a Taiwan Strait contingency.