The open-source large model released by the Chinese artificial intelligence (AI) company DeepSeek has taken the world by storm over the past week, sending shockwaves not only through the U.S. tech industry but also reaching the Pentagon, which has invested heavily in AI. Multiple U.S. media outlets have noted the contradictory stance of the U.S. military toward DeepSeek: while there are concerns about so-called “personal data leaks,” there is also excitement about the new path for AI development that DeepSeek has introduced.
U.S. Military Issues Emergency Ban on DeepSeek
According to the U.S. defense news website Defense One, China’s DeepSeek has launched an open-source generative AI model that rivals top U.S. AI models while requiring only a fraction of the time and financial resources that companies like OpenAI invest in training their models. DeepSeek’s breakthrough has triggered a wave of concern across the White House, Wall Street, and Silicon Valley. Former U.S. President Donald Trump even declared, “This is a wake-up call for our industry—we need to focus on competition with China.”
Reports indicate that benchmark testing shows DeepSeek’s model is highly competitive in reasoning-intensive tasks, consistently achieving top-tier performance in mathematics and coding. However, in non-reasoning tasks and factual accuracy, it still lags behind OpenAI’s most advanced models.
Despite these differences, DeepSeek’s ease of use and low cost have won praise from various sectors in the U.S., including Pentagon staff. According to Bloomberg, U.S. military personnel had already started downloading earlier versions of DeepSeek’s code onto their work computers as early as the fall of 2024. In response to the AI model’s soaring popularity, the U.S. Navy was the first to issue a ban, followed by the Defense Information Systems Agency, which prohibited its use within military networks.
The Pentagon justified the ban by citing so-called “security and ethical concerns” related to DeepSeek’s origins and applications. Defense One reported that U.S. military officials fear that widespread personal data leaks in the U.S. represent a critical “national-level vulnerability,” which adversaries could exploit in the event of a conflict. The rapid emergence of highly capable AI models like DeepSeek, they argue, could exacerbate this risk.
In response to the U.S. ban and related measures, China’s Permanent Representative to the United Nations, Fu Cong, addressed the issue during a press conference at the UN headquarters in New York on February 4 (Beijing time). He stated, “Never underestimate the intelligence and ingenuity of Chinese researchers. The global sensation and anxiety surrounding DeepSeek prove that technological containment and restrictions do not work. This is a lesson the world, particularly the United States, needs to learn.”
Fu emphasized that rather than imposing more bans, the U.S. and China—two leading nations in AI—”cannot afford not to cooperate.” He stressed that only through collaboration could both countries bridge the digital and intelligence divide, ensuring that developing nations also benefit equally from AI advancements.
Impact on Mobile Nuclear Reactor Projects
While the U.S. military has cited “data security concerns” to justify restrictions on DeepSeek, Defense One noted that the AI model’s novel technological approach offers significant advantages for military operations in remote areas with limited or unstable internet access. This capability could provide valuable AI-powered tools to troops in such environments. Additionally, DeepSeek’s efficiency and low energy consumption could benefit the Pentagon, which seeks to maximize AI capabilities while minimizing costs.
On February 4, Defense News highlighted that the U.S. military has been heavily investing in AI to enhance deployment efficiency. For instance, a revised aviation strategy released by the U.S. Marine Corps on February 3 prioritizes the use of autonomous systems, drones, and AI-powered software “to ensure the survival of its fighter fleet in high-intensity conflict zones.” However, Pentagon officials have warned that the U.S. lacks the energy resources and computational power needed to sustain large-scale AI infrastructure, making it difficult to fully implement these technologies.
The latest report noted that the massive power and infrastructure demands of mainstream U.S. AI models were evident in the ambitious “Stargate Project,” promoted during the Trump administration. Recently, tech giants such as OpenAI, SoftBank, and Oracle announced plans to jointly invest $500 billion in building new AI infrastructure across the U.S., aiming to maintain America’s leadership in global AI competition.
Roy Campbell, Associate Director for Advanced Computing at the Pentagon’s Office of the Under Secretary of Defense for Research and Engineering, acknowledged that many U.S. overseas military bases lack the computing power necessary for training AI models. He explained that in some cases, data must be transmitted back to the U.S. for processing at the Department of Defense’s supercomputing centers—a practice that significantly reduces efficiency and imposes high demands on stable communication.
To address this issue, Defense News revealed that one of the Pentagon’s proposed solutions is to develop mobile nuclear reactors to power AI systems at frontline bases. Jeff Waxman, who oversees the Pentagon’s Strategic Capabilities Office project on mobile nuclear reactors, stated that the U.S. military launched its “Portable Nuclear Reactor” initiative in 2019 to mitigate the immense energy strain imposed by AI and high-power computing technologies.
However, the emergence of DeepSeek—a model capable of rivaling top U.S. AI systems while consuming significantly less power and requiring fewer chips—has raised concerns among Western analysts. If such technology becomes widely adopted, it could reduce the overall energy demand for AI applications, potentially undermining the necessity and feasibility of the Pentagon’s mobile nuclear reactor project. Given the security risks associated with deploying portable nuclear reactors in conflict zones, analysts speculate that this project’s future may now be in jeopardy.
“The U.S. Military Doesn’t Need General-Purpose AI Models”
Defense One also pointed out that DeepSeek’s breakthrough presents an opportunity to develop more efficient tools, which could benefit the U.S. military as well. The prevailing “bigger is better” approach in U.S. AI development—focusing on acquiring more training data, building larger models, and constructing massive data centers—is actually crowding out the edge computing projects that the military truly needs.
The report states that the performance of large U.S. AI models does not necessarily justify the vast resources required to build and sustain them. The U.S. Department of Defense is developing AI along two technological paths: large-scale models that demand extensive computing resources and lightweight AI capable of running on small platforms even when disconnected from networks. For a long time, U.S. researchers have primarily focused on general-purpose AI models, neglecting smaller, specialized AI models tailored for specific fields. However, in actual military operations, the U.S. military does not require the complex functionalities of general AI models, as its primary needs are concentrated in a few specialized areas.
The report provides examples of various AI-driven needs that frontline troops may encounter, such as using AI to filter specific types of vehicles from drone or satellite imagery, analyzing certain electromagnetic signal characteristics they come across, or even simply gathering local economic, weather, population, or consumer data to plan more effective and safer operations in dense urban environments. In these scenarios, the U.S. military requires AI models that can operate on relatively small datasets, with computing power that does not rely on large-scale servers or GPUs. On the contrary, due to the unpredictable conditions in frontline areas, which may involve enemy electronic interference, limited power supply, and weak communication infrastructure in small forward bases, a highly efficient tool like DeepSeek—requiring relatively minimal computing power and energy—actually aligns better with the Pentagon’s needs.
Defense News mentions that another potential solution to the AI capability gap is improving processor efficiency. Steven Meier, Deputy Director for Space Technology at the U.S. Naval Research Laboratory, stated that his team is exploring the use of more efficient neuromorphic processors, which can be 100 times more efficient than standard processors. He explained, “Neuromorphic processors take up less space, operate faster, and consume significantly less power.”
Leave a comment