DeepSeek R1: The “Impressive Model” & Military Affairs

A quick assessment of the potential consequences for military affairs of the new approach to AI demonstrated by DeepSeek-R1. We still have much to learn.

The past 24 hours have seen technology and business publications reporting on the release of the DeepSeek-R1 chatbot in the United States. The developers of the latest version of the DeepSeek AI model have claimed that it operates on par with OpenAI-o1, that it is fully open-source and that it cost just $6 million to develop. OpenAI CEO Sam Altman described it as “an impressive model”.

Understandably, investors in the United States have questioned the billions of dollars they have been pouring into American developers. Chip maker Nvidia lost $588 billion in value in a single day, which is slightly more than the GDP of Norway. Besides the reassessment of tech investment in America, and the losses suffered by tech companies, there may be a range of technological, commercial and political implications of the new DeepSeek-R1 model. As one commentator has noted, “it upends the way that investors have thought about how AI needed to be developed and implemented.”

I intend to explore in this article the potential military implications of this DeepSeek-R1 development. I should emphasise that I write this as an expert in military affairs, and someone who has written about the impact of new technologies on military organisations, not as someone who is an AI or software expert. Despite that, I think there are sufficient implications that we can identify now – and many questions that should be asked.

What Is DeepSeek-R1?

DeepSeek is a Chinese artificial intelligence company founded in July 2023 in the Chinese city of Hangzhou. It released its first DeepSeek LLM in November 2023, with DeepSeek-V2 released in May 2024 and DeepSeek-V3 released in December 2024.

According to a report by the BBC, DeepSeek was founded by information and electronic engineering graduate, Liang Wenfeng, who was able to amass tens of thousands of Nvidia A100 chips, which are now banned from being exported to China. A report from Scientific American has proposed that DeepSeek engineers could have used those Nvidia A100 chips to develop the latest R1 model. But the crucial breakthrough appears to be that DeepSeek then used lower-powered Nvidia H800 chips to train its new model.

The DeepSeek-R1 AI-powered chatbot has become the most downloaded free app on Apple's app store. It was claimed in a Scientific American article that, “the new model has 670 billion parameters, or variables it learns from during training, making it the largest open-source large language model yet. But the model uses an architecture called mixture of experts so that only a relevant fraction of these parameters—tens of billions instead of hundreds of billions—are activated for any given query. This cuts down on computing costs.”

DeepSeek-R1: Some Unknowns

Given this chatbot has only been released in the past few days, it is yet to be exposed to the global testbed of technical experts and everyday users of AI in society. And, it has already imposed restrictions on sign up, due to alleged cyber-attacks. Could this be also masking a problem with scaling up the chat bot for widespread use?

Cost is another unknown. While DeepSeek has claimed this version of its chatbot only cost $6 million to develop, what range of costs are included and excluded from this figure. The $6 million development figure is orders of magnitude less than what Open AI and other models cost to develop. While this isn’t to say the developers of DeepSeek are lying, we might exercise some sensible scepticism on this low-cost AI until more data is available. And where did the money come from? Was there any PLA or CCP funding used to develop the model and if so, what does this mean about its reliability?

Images: @AndrewBGreene

Censorship already appears to be an issue. Initial user query’s about Chinese President Xi’s policies and the 1989 Tiananmen Square protests and massacre have mimicked Sergeant Schultz from Hogan’s Heroes – “I know nothing!” These are known censorship issues. What else has been deliberately left out from DeepSeek-R1 training to protect the developers from a harsh response from the CCP?

Given these unknowns, what are the implications for military institutions?