The release of the DeepSeek-R1 chatbot in the English will have sparked the interest of military and intelligence agencies “across the Western World,” according to one analyst.
Mick Ryan is a retired Australian Army Major General who is now senior fellow for military studies at the Sydney-based Lowy Institute and the Australia Chair for the Center for Strategic and International Studies.
“I expect that there will be a rush for military and intelligence agencies across the Western world to download and test it extensively,” he says, though he adds there will also be “many questions ... about its effectiveness compared to currently available models, and about how it might be better for battlefield applications where power and efficiency are more significant concerns than in static, strategic headquarters.”
Artificial intelligence (AI) is already in use in military applications and has come to the fore in the Russia-Ukraine conflict, he says.
“In some respects, Ukraine has acted as an AI testbed for this technology, and many other military organisations are either introducing it or plan to introduce it for command and control, speeding up targeting, imagery analysis, logistics, personnel tracking and a range of other missions,” Ryan says.
“The key issue is this: with existing AI models, their cost and required computing power meant that we could not afford to do everything we wanted to with AI. Might [AI that uses] the same approach as DeekSeek-R1 mean military organisations can do everything they want to do with AI, at the levels they want to do it, in a much quicker timescale than imagined?”
Obviously, though, there are major security concerns with adopting a technology developed under the supervision of the Chinese Communist Party (CCP).
“Given this new AI model has emerged from China, one of the first order concerns will be about how data that is generated by users is sent back to China, and how it might be used by the CCP’s various intelligence agencies or the [Chinese military],” Ryan says.
“It makes sense to be cautious, and I certainly won’t be downloading it anytime soon. And I doubt anyone with a security clearance in any western military or intelligence service will be either, outside of official applications.”
It’s a “remote possibility” that Beijing’s military has adopted DeepSeek technology, Ryan believes.
“If they do, have they been able to incorporate it into weapon systems as well as targeting and decision-making at the strategic, operational and tactical levels? If they have been able to do this, what does it mean for the correlation of military forces in the Pacific?” he asks.
Dubious Cost
DeepSeek is a Chinese artificial intelligence company founded in July 2023 in Hangzhou. It released its first DeepSeek LLM (large language model) in November 2023, followed by DeepSeek-V2 in May 2024 and DeepSeek-V3 in December 2024.But it is the recent availability of an English language version that has seen it grab headlines across the West.
The company’s founder, Liang Wenfeng, claims it was developed for $6 million—a figure that has been taken at face value by many commentators, including U.S. President Donald Trump, who characterised it as a “wake-up call” for American developers.
Chinese AI researchers are severely disadvantaged, compared to their Western counterparts, by the hardware available to them due to export restrictions on chips.
DeepSeek is believed to have used some 2,048 of Nvidia’s H800, a slower version of the H100 GPU designed specifically for the Chinese market before the Biden administration closed the loophole that allowed chipmakers to export slower units.
However, the $6 million figure focuses largely on training costs for DeepSeek-V3, GPU rental costs, and the processing of around 14 trillion tokens.
It excludes the cost of developing the two earlier versions and any failed iterations of the model, as well as the costs of data gathering and corpus creation (a corpus is the material the AI reviews to become intelligent in whatever it was designed for), researcher salaries, actual hardware costs, and infrastructure considerations like cooling and maintenance.
Martin Vechev, a professor at ETH Zürich and founder of the Institute for Computer Science, Artificial Intelligence and Technology (INSAIT) in Bulgaria, said via LinkedIn that the numbers were “somewhat misleading.”
“The real price to develop such a model is many times higher than $5 to 6 million,” he wrote.
“Further, 2,048 H800 [chips] costs between $50 and 100 million, and the company that owns DeepSeek is a large Chinese hedge fund, which has much more than 2,048 H800s.”
These doubts won’t stop military institutions from asking, “Why should we ever pay big American AI companies what we have been paying them for AI services, ever again?” Ryan says, adding that they should be asking that question.
