In February, China’s state-owned defense powerhouse, Norinco, showcased an innovative military vehicle designed for autonomous combat-support operations, capable of speeds up to 50 kilometers per hour. This groundbreaking machine is powered by DeepSeek, an artificial intelligence model that stands as a testament to China’s burgeoning technological prowess.
The introduction of Norinco’s P60 vehicle was hailed by Communist Party officials as a clear signal of Beijing’s strategic intent to integrate DeepSeek and advanced AI into its military. This move highlights China’s aggressive efforts to close the technological gap with the United States in the ongoing arms race, as both nations intensify their military preparedness for potential future conflicts.
A comprehensive review by Reuters, examining countless research papers, patents, and procurement documents, reveals Beijing’s meticulously planned initiative to exploit artificial intelligence for significant military advantages.
While the precise mechanisms and deployment scale of China’s next-generation AI-powered weaponry remain highly classified, public procurement records and patent filings provide intriguing insights. These documents suggest substantial progress by Beijing in areas like autonomous target identification and real-time battlefield decision-making, closely paralleling developments in U.S. military AI research.
It’s important to note that Reuters could not independently verify the full-scale production of all these military technologies, and patents do not always translate directly into operational systems.
Despite U.S. export controls, the People’s Liberation Army (PLA) and its affiliated organizations continue to seek and utilize Nvidia chips, including models subject to restrictions, as evidenced by reviewed documents, tenders, and patents.
The exact timeline of these acquisitions remains unclear; Reuters was unable to confirm if the chips were procured before Washington’s restrictions took effect. However, recent patent filings, some as late as June, confirm their ongoing use by Chinese military research institutions. Notably, the U.S. Commerce Department imposed a ban on exports of Nvidia’s advanced A100 and H100 chips to China in September 2022.
Nvidia spokesperson John Rizzo addressed the situation, stating to Reuters that while tracking individual resales of older products is not feasible, “recycling small quantities of old, second-hand products doesn’t lead to new capabilities or national security concerns.” He emphasized that employing restricted products for military uses would be impractical without continuous support, software updates, or maintenance.
Both the U.S. Treasury and Commerce Department have yet to respond to inquiries regarding Reuters’ findings.
In a significant shift, the Chinese military has expanded its engagement with contractors who reportedly rely solely on domestically produced hardware, such as Huawei AI chips. This observation comes from Sunny Cheung, a fellow at the Jamestown Foundation, a Washington-based defense policy think-tank, who scrutinized hundreds of tenders from the PLA Procurement Network over a six-month period this year.
While Reuters could not independently corroborate Cheung’s claim, such a transition would align with Beijing’s ongoing public campaign urging domestic companies to prioritize and utilize Chinese-made technology.
Reuters’ own examination of procurement notices and patents submitted to China’s patent office confirmed a demand for and utilization of Huawei chips by PLA-affiliated entities. However, the news agency was unable to verify every tender documented by the Jamestown Foundation, which shared its upcoming report with Reuters in advance.
Huawei remained silent when questioned about the military deployment of its chips. The Chinese defense ministry, DeepSeek, Norinco, and the various universities and defense firms involved in the reviewed patents and research papers all declined to comment on their engagement with AI for military purposes.
DeepSeek models were prominently featured in approximately a dozen tenders from PLA entities reviewed by Reuters this year, significantly outnumbering the single reference to Alibaba’s rival AI model, Qwen.
Alibaba did not respond to requests for comment regarding the military’s potential use of its Qwen AI.
According to the Jamestown Foundation, procurement notices related to DeepSeek have shown a marked increase throughout 2025, with a steady stream of new military applications surfacing on the PLA network.
DeepSeek’s growing appeal within the PLA also underscores China’s broader objective of “algorithmic sovereignty” – a strategic push to minimize reliance on Western technology and solidify its command over vital digital infrastructure.
The U.S. Department of Defense declined to offer comments on the PLA’s utilization of AI.
The documents indicate that China is actively exploring advanced AI applications, including robot dogs capable of scouting in coordinated packs, drone swarms designed for autonomous target tracking, and the development of highly immersive command centers and sophisticated war game simulations.
Illustrating this ambition, in November 2024, the PLA issued a tender for AI-powered robot dogs, envisioned to collaboratively scout for threats and neutralize explosive hazards, echoing concepts typically found in science fiction.
While Reuters could not confirm whether this specific tender was ultimately fulfilled, state media reports have previously shown China deploying armed robot dogs, manufactured by AI robotics company Unitree, in military exercises.
Unitree did not respond to inquiries concerning its involvement with the PLA.
An analysis of patents, tenders, and research papers from the last two years clearly demonstrates the PLA and its associated organizations’ commitment to using AI for enhanced military planning. This includes the development of technology for rapid analysis of satellite and drone imagery.
Landship Information Technology, a Chinese firm specializing in integrating AI into military vehicles like those from Norinco, highlighted its capabilities in a February white paper. The company claimed its Huawei-chip-powered technology could swiftly identify targets from satellite imagery and coordinate seamlessly with radars and aircraft for operational execution.
Xi’an Technological University reported that AI has dramatically reduced the critical time frame between target identification and operational execution for military planners.
In May, researchers from the institute published findings indicating that their DeepSeek-powered system could analyze an astonishing 10,000 diverse battlefield scenarios, each with unique variables, terrain, and troop deployments, in a mere 48 seconds.
They further asserted that a comparable task would typically require a conventional team of military strategists approximately 48 hours to accomplish.
Reuters was unable to independently verify these specific claims made by the researchers.
The reviewed documents strongly suggest a significant investment by Chinese military entities in developing increasingly autonomous battlefield technologies.
Evidence from two dozen tenders and patents reviewed by Reuters reveals the military’s ongoing efforts to integrate AI into drones. This integration aims to enable drones to autonomously recognize and track targets, and to operate in coordinated formations with minimal human oversight.
According to a patent filing this year, Beihang University, a leading institution in military aviation research, is leveraging DeepSeek to enhance the decision-making capabilities of drone swarms. The focus is on engaging “low, slow, small” threats, which is military jargon for other drones and light aircraft.
Amidst rising concerns that a potential conflict between Beijing and Washington could result in the unconstrained use of AI-powered weaponry, Chinese defense leaders have publicly affirmed their commitment to preserving human control over such systems.
Meanwhile, the U.S. military, also a significant investor in AI, plans to deploy thousands of autonomous drones by late 2025. This initiative, according to officials, is a direct response to counter China’s perceived numerical superiority in unmanned aerial vehicles.
Chinese defense contractors, such as Shanxi 100 Trust Information Technology, have actively promoted their use of domestically manufactured components, including Huawei’s Ascend chips, which are essential for operating advanced AI models.
The company did not respond to inquiries regarding its connections with Huawei and the PLA.
Despite the push towards domestic processors, Nvidia hardware is still frequently referenced in research by military-affiliated academics, as revealed by a review of patent filings from the last two years.
Reuters identified 35 applications that cited the use of Nvidia’s A100 chips by academics at prominent Chinese institutions. These include the PLA’s National University of Defense Technology (NUDT) and the “Seven Sons” universities, a group under U.S. sanctions known for their significant defense-related research contributions to Beijing.
During the same period, these entities also filed 15 patents for AI applications that specifically mentioned Huawei Ascend hardware, originally developed as an alternative to Nvidia chips.
Notably, in June, the PLA Rocket Force University of Engineering submitted a separate patent for a remote-sensing target detection system, explicitly stating its reliance on A100 chips for model training.
Last year, Senior Col. Zhu Qichao, head of a NUDT research center, acknowledged to Reuters that U.S. restrictions have “to some degree” affected their AI research, but he affirmed their unwavering determination to close the technological gap.
Nvidia’s Rizzo, however, downplayed the PLA’s demand for Nvidia hardware, suggesting that China “possesses ample domestic chips to fulfill all its military application requirements.”