From Drone Swarms to Robot Dogs — DeepSeek Becomes China's Secret Weapon in AI Arms Race

The global power play of the 21st century isn't fought with tanks and missiles, but with lines of code. The AI arms race is on, and the ultimate prize is supremacy. While the world watches the West's most prominent players, a powerful, quieter contender is emerging from the East.

Suddenly, DeepSeek is being deployed for tasks ranging from coordinating autonomous drone swarms to controlling next-generation robot dogs, quietly transforming into China's secret weapon in this high-tech, high-stakes conflict.

China Weaponises AI for Arms Race

In February, China's state-owned defence giant, Norinco, revealed a military vehicle designed to autonomously perform combat-support duties at speeds of 50 kilometres per hour. The system's operational intelligence drew its power from DeepSeek, the firm whose artificial intelligence model is widely celebrated across China's technology industry.

Communist Party officials quickly promoted the Norinco P60's debut in public announcements, calling it an initial demonstration of how Beijing employs this advanced AI to gain ground in the arms competition against the US.

This demonstration comes at a time when political figures in both nations are pressing their defence forces to ready themselves for potential hostilities.

A Reuters analysis of hundreds of research documents, patents, and purchase orders provides a clear picture of Beijing's organised campaign to leverage artificial intelligence for military superiority.

The Hardware Conflict: Nvidia Use Persists

The exact operational details of China's next-generation weapon systems, as well as the scope of their deployment, remain state secrets. However, purchase documents and patent filings provide insight into Beijing's advancements in areas such as autonomous target identification and immediate combat decision support, efforts that closely resemble those being undertaken by the US.

Crucially, Reuters did specify, however, that it was unable to confirm whether every one of these items had been constructed, and that patent approvals do not always translate into fully functioning technology.

Yet, the People's Liberation Army (PLA) and its affiliated organisations persistently seek and utilise Nvidia microchips, including specific models subject to US export restrictions, as evidenced by these documents, purchase orders, and patent applications.

Reuters was unable to determine whether the microchips had been accumulated before Washington introduced controls, as the records fail to specify when the hardware was originally shipped. Patent applications submitted as late as June, however, are still confirmed by research facilities connected to the military.

This usage comes despite the US Commerce Department's ban on exporting Nvidia's sought-after A100 and H100 chips to China, a prohibition that began in September 2022.

Nvidia's Defense and the Domestic Pivot

Nvidia spokesperson John Rizzo, in a comment given to Reuters, stated that while the company cannot trace every individual resale of goods sold earlier, he maintained: 'recycling small quantities of old, second-hand products doesn't enable anything new or raise any national security concern. Using restricted products for military applications would be a nonstarter without support, software, or maintenance.'

Sunny Cheung, a fellow at the Washington-based Jamestown Foundation defence policy think-tank, noted that the Chinese military also stepped up its use of suppliers claiming to rely solely on domestically produced hardware, such as Huawei's AI chips.

Cheung reached this conclusion after examining several hundred purchase requests released by the PLA Procurement Network during a six-month period this year. Reuters was unable to verify his claim independently, but this apparent change would align perfectly with Beijing's public drive to compel domestic companies to adopt Chinese-developed technology.

Huawei Use Confirmed Amid Universal Silence

The news agency's examination of purchase orders and patents submitted to China's patent office confirmed that PLA-linked groups were both seeking and using Huawei microchips. It could not, however, authenticate every tender seen by Jamestown, which is due to publish its report this week after providing an early copy to Reuters.

Huawei refused to comment on enquiries regarding the military use of its microchips. Similarly, the Chinese Defence Ministry, DeepSeek, and Norinco failed to acknowledge requests for details on their deployment of AI for combat purposes.

Furthermore, the academic institutions and defence companies responsible for the patents and research documents viewed by Reuters offered no response to similar questions.

PLA Ramps Up DeepSeek Use

DeepSeek models were featured in a dozen PLA purchase requests this year, whereas Alibaba's rival model, Qwen, was mentioned just once. These purchase notifications for DeepSeek have steadily increased throughout 2025, which underscores China's drive for 'algorithmic sovereignty' — a strategy aimed at reducing its dependence on Western technology.

The US Department of Defence (DoD) declined to comment. Still, a State Department spokesperson asserted that 'DeepSeek has willingly provided, and will likely continue to provide, support to China's military and intelligence operations'.

The spokesperson further indicated that Washington would 'pursue a bold, inclusive strategy to American AI technology with trusted foreign countries around the world, while keeping the technology out of the hands of our adversaries'.

Washington's Stance

China is actively investigating the use of AI-powered robot dogs for group scouting, self-operating drone swarms, visually immersive command facilities, and sophisticated war-game exercises. Separately, in November 2024, the PLA published a purchase request for robot dogs intended to identify dangers and dispose of explosive hazards, although whether this order was ever fulfilled remains unclear.

Unitree, a firm that has previously fielded armed robot dogs during exercises, did not provide a comment. Furthermore, PLA research conducted over the last two years indicates that artificial intelligence is now being leveraged for rapid analysis of satellite and drone imagery, as well as for coordinating battlefield operations.

Landship Information Technology claimed its Huawei-microchip-based AI has the capability to pinpoint targets and synchronise with both radar and aircraft. Further illustrating AI speed, Xi'an Technological University reported that its DeepSeek-driven system evaluated 10,000 battlefield situations in just 48 seconds — a task that would take 48 hours for human planners — but Reuters was unable to independently confirm this claim.

AI Powers PLA's Drone Swarm Future

Military organisations across China are committing significant capital to autonomous combat technologies. Approximately 24 purchase requests and patent filings examined by Reuters indicate a drive to incorporate artificial intelligence into uncrewed aerial vehicles (UAVs) capable of identifying, following, and operating in coordinated groups with very little human supervision.

Beihang University is employing DeepSeek to sharpen swarm decision-making capabilities, particularly against "low, slow, small" targets like uncrewed aerial vehicles and light aircraft.

While Chinese defence chiefs stress the importance of maintaining human oversight for weapon systems, the US military is concurrently increasing its AI spending, with plans to field thousands of autonomous drones by the close of 2025 to offset China's expanding UAV superiority.

Hardware Dynamics and PLA Strategy

Chinese defence contractors, including Shanxi 100 Trust, have publicly advertised their reliance on homegrown components, such as Huawei's Ascend microchips, though they have refused to discuss any affiliations with the PLA. Nvidia hardware, however, continues to see extensive deployment.

Reuters discovered 35 filings referencing the Nvidia A100 chip by researchers associated with the military at the National University of Defense Technology and the 'Seven Sons' universities, compared to 15 patents that cited Huawei Ascend as an alternative.

As recently as June, the PLA Rocket Force University submitted a patent application describing the use of A100 microchips to train a remote-sensing model. Senior Col. Zhu Qichao conceded that US restrictions have impacted research 'to some degree', but stressed that initiatives to reduce the capability gap are ongoing.

Contrasting this, Nvidia's Rizzo downplayed demand, asserting that China 'has more than enough domestic chips for all of its military applications'.