[Editor’s Note: Army Mad Scientist is pleased to feature a co-effort by returning author, and current Army Mad Scientist Intern, Wyatt Forney and Mad Scientist Laboratory Editor, Kunal Chauhan. In their post, the team brings to light the growing power of Chinese artificial intelligence models, the advantage of their open-source nature, and how this empowers their military approach to total systems warfare! — Read on!]
The Open-Source Offensive
In the AI race, much of the focus and effort so far has gone to understanding the way China is trying to close the gap with the United States and how the U.S. could stay ahead. Less focus has been given to the way China is actually developing its AI models and the current capabilities and advantages that are provided from them. While the capability gap is closing, it is worth looking into how China views AI institutionally, and how they intend to use and distribute its power. There are pertinent military applications here as well, with Chinese models crossing dangerous thresholds into autonomous cyber operations and agentic coding, which can greatly enhance the People’s Liberation Army’s (PLA) operational capacity.
THE CLOSING GAP
While American models generally are more powerful, they are also far more closed off to outside actors. China takes advantage of this by creating open-source AI models, like DeepSeek, which took the West by storm, since it almost matched the capability of leading American models without the same level of computational power. These models are far cheaper for developer use, and in the case of DeepSeek, completely free for users, which makes it attractive to those who want powerful AI models without the cost of the latest OpenAI GPT model. Large American companies have also begun to use these Chinese models, though there is no accurate data to the extent of use since it is not publicly available information. Anyone can access the models, customize them to fit their needs, and then reupload them publicly to AI code repositories like Hugging Face. This could give China an edge in the informational warfare space, gaining Global South support by characterizing themselves as providing AI infrastructure at a low cost while the hegemonic US hoards its technology.
Because its models are open-source and constantly being integrated into government, private sector, and military projects, China is able to accelerate their development through a constant feedback loop, where larger more powerful models can train smaller, student models. Through this sort of distillation practice, China is able to build power for their models without the same level of compute power that American models have in their closed system. Essentially, they are able to do more with less due to their open ecosystem which creates constant refinement. It is important to note however, that much of this distillation is forced by scarcity, specifically US export controls on semiconductors that prevent China from having the same level of computational capabilities as the US. On the cybersecurity side, China is better poised to deal with threats to its models since its Cyberspace Administration of China (CAC) is deeply entrenched in data flows and model development. Due to the authoritarian nature of the government, they are simply able to tell companies what to do and what to turn on and turn off, which could make them able to deal with threats quicker.
Chinese models are also crossing critical thresholds in being able to conduct cyber operations, such as the recently released model from Z.ai, which features advanced agentic coding capabilities. This can unlock dangerous operational utility for the PLA and Chinese intelligence services, especially with an AI that can operate somewhat autonomously.
GLOBAL INFLUENCE
As the capability gap closes between American and Chinese AI models, corporations now have a choice to make between affordability and quality. Chinese open-source AI models are cheap, highly deployable, and customizable to a business’s needs, and American companies are beginning to go with options like DeepSeek and Qwen over models from OpenAI and Anthropic, especially for repetitive tasks. Using a top-of-the-line American model just to write lines of code creates a price mismatch that companies must address.
Businesses as large as Airbnb have announced usage of Chinese models to cut costs, but their transparency is the exception, not the rule. Companies are in no way required to announce which models they use and for what tasks, making Chinese model reliance difficult to track. This reliance could leave corporations vulnerable to hidden backdoors in the code that could be used to steal or manipulate information. 
China recently founded the World Artificial Intelligence Cooperation Organization (WAICO), uniting twenty-nine countries with the common goal of collaborative AI development. Of the countries on this list, all but Brazil are official members of China’s development-focused Belt and Road Initiative (BRI). Brazil, like other members, has previously collaborated with China in BRI-adjacent development deals. WAICO is a powerful tool to further impress China’s developmental footprint onto partnered countries by introducing Chinese AI model reliance across the Global South. Some of its member countries already have a disproportionate market share of DeepSeek, most notably Cuba, Belarus, and Russia.
Surveyed populations in Southeast Asian countries like Indonesia and Malaysia report markedly high trust in governmental ability to regulate AI and optimism about AI’s potential impact on their quality of life, making China’s AI diplomacy more impactful in this region. These relationships will undoubtedly add to the national security vulnerabilities many countries in the region already face as a result of lopsided economic policies with China. The countries also represent additional sandboxes in which China can test its AI models, as the open-source nature of the models adds to their utility in the hands of a country’s government attempting to expand its infrastructure and economy. This feedback loop, where open-source AI is trained on new data through use in novel situations, will be another boost to the capabilities of Chinese AI models in an environment where American models have a much weaker foothold.
Another geopolitical concern emerges from the Global South’s reliance on Chinese AI models. As countries expand their total use of AI and look to expand their own development, they will need to increase their compute capacity through the construction of data infrastructure. China could easily cash in on a vertical integration scheme by looking to provide the data centers as well as the AI systems they harness, further increasing their cultural and economic influence in countries across the southern hemisphere. This relationship would look similar to China’s previous engagements with some WAICO member countries like Brazil and Serbia regarding surveillance system exports, where China was able to provide both the necessary software and hardware and turn a greater profit. 
Further stalls in the development of high-quality American AI models could put Chinese models on even footing in terms of capabilities, with their importance to the developing economies of the Global South and affordability for American corporations threatening both international influence and US national security. Due to the distillation system that China utilizes and the reality of compute scarcity that they face, they have been able to develop models that both nearly match the capability of more powerful US models and require less supportive infrastructure. This means that the PLA will have access to similar capabilities with far less cost, which could allow investment into other pertinent military areas and enhance their operational capacity.
TOTAL SYSTEMS WARFARE
The strategy of the PLA emphasizes the importance of systems destruction against enemies attempting to intervene in its military operations. Systems destruction warfare means the confrontation between two complete military systems, where eliminating an enemy’s operational capabilities means targeting their information capabilities as well. The PLA’s growing adoption of the concept of “intelligentization” seeks to fuse this approach with the implementation of AI agents that can algorithmically improve the PLA’s offense and defense across multiple fronts, harnessing their own data while identifying weaknesses in an enemy’s cyber infrastructures to cripple their forces.
The “cognitive domain” is an area that China will seek to control in the event of a conflict with the United States. Restricting available information or pushing disinformation is another component of the AI-oriented total systems warfare that the PLA will seek with the United States. Through these propaganda and disinformation campaigns, China hopes to destabilize retaliation efforts and sow disunity in the American public. Social media platforms will be the frontlines for this effort, with algorithms that can be targeted by hostile actors to push or suppress specific content. Most platforms also lack proper filters or tags for AI-generated content. Some, like X (formerly Twitter), even tout their own top-of-the-line generative AI capabilities. Already, generative AI can create videos nearly indistinguishable from reality. Further years of development backed by the malign intent of a hostile state actor will surely enhance its capabilities for mass confusion. 
Although AI’s military impacts can be easily understood in the domains of cyber and electronic warfare, its role in training the forces behind the PLA’s kinetic strikes is also substantial. Because the PLA has not received real combat experience for the last fifty years, AI is one of the best options it has at its disposal to design relevant training exercises for its soldiers. These simulations give the PLA’s commanders insights into battlefield scenarios they have never participated in themselves and are key to their strategy design. The PLA also looks to use AI for operational decision-making, a practice antithetical to Western militaries with human input at each operational level. The PLA views this incorporation of AI as an optimization of their human capabilities and a scientific approach to total victory.
The United States still has the AI capability advantage, but the limited deployability of its own models hamstrings their utilization at each level of military operation. As the gap between our models and China’s continues to shrink, the next step could be the adoption of open-source, collaboration-oriented models that can be modified to fit a team’s needs and can train on the fly. The US Army does not need to sacrifice its human approach to decision-making to further integrate AI into its operations, but it may need to learn some new tricks to adapt to the total systems warfare approach the PLA will use.
If you enjoyed this post, check out the T2COM G-2’s Operational Environment Enterprise web page, brimming with authoritative information on the Operational Environment and how our adversaries fight.
About the Authors:
Wyatt Forney is an intern with the Army Mad Scientist team as a part of T2COM G-2. He is currently a third-year student at the University of Richmond studying Political Science and Linguistics, with a focus in Middle Eastern politics and Modern Standard Arabic (الفصحى). He looks to pursue a career in intelligence after graduation. He also enjoys being a radio DJ at the University of Richmond.
Kunal Chauhan is an intelligence analyst with T2COM G-2 and prior Mad Scientist intern. He has prior defense experience interning with the US Army War College and conducting research through the Defense Innovation Unit. He recently received his BA in International Relations from William & Mary and is interested in emerging technology in warfare, defense procurement and legislative affairs. In his free time Mr. Chauhan enjoys traveling, biking, and reviewing new restaurants.
Disclaimer: The views expressed in this blog post do not necessarily reflect those of the U.S. Department of Defense, Department of the Army, or the Transformation and Training Command (T2COM).
SOURCES:
https://www.brookings.edu/articles/china-is-running-multiple-ai-races/
https://hai.stanford.edu/ai-index/2026-ai-index-report
https://www.npr.org/2026/07/15/nx-s1-5886476/startups-cheap-chinese-ai-models
https://greyjournal.net/hustle/grow/should-startups-use-chinese-ai-models/
