[Exclusive] "Only 1% of Needed GPUs Secured, No Place to Store Them"... AI Defense Plan to Cut Guard Personnel by 16,000 in Doubt

Ministry of National Defense: "Reducing GOP Personnel and Transitioning to AI Surveillance System"
Severe Shortage of Key AI Component GPUs
Lack of Large-Scale Storage Centers Even if Additional GPUs Are Secured

The Ministry of National Defense currently possesses only about 1% of the graphics processing units (GPUs) required for its needs. Although the ministry has emphasized national defense security leveraging artificial intelligence (AI), it is significantly lacking in the number of GPUs, which serve as the “brain” of such systems. Furthermore, even if additional GPUs were acquired, there is not even a center available to store them.

The Ministry of National Defense plans to transition the frontline General Outpost (GOP) to an AI-based scientific system, reducing the number of guard personnel to approximately 6,000.

According to the Ministry of National Defense on April 28, the military plans to gradually reduce the number of guard personnel at the frontline General Outpost (GOP) and transition to an AI-based scientific monitoring system. On April 7, Minister of National Defense Ahn Gyu-baek stated at a press briefing, “Currently, there are about 22,000 guard personnel stationed at the frontline GOP, but with the implementation of an AI-based scientific system, this number will be reduced to approximately 6,000.”

The military estimates it will need about 50,000 GPUs to immediately realize AI capabilities. In contrast, the number of GPUs currently available at the Ministry of National Defense’s data centers is extremely limited. Even when combining the holdings at the national defense data centers and the Agency for Defense Development (ADD), the total amounts to only around 500 GPUs—just about 1% of the requirement.

The Defense Integrated Data Center, which manages the defense network, has 32 GPUs; the IT rooms of each military branch have 55 GPUs; and the Battlefield Data Center under the Defense Communication Command, responsible for communications during wartime, has only 4 GPUs. The Agency for Defense Development, which researches weapon systems, also possesses 480 GPUs. Although it planned to secure an additional 320 units last year, the plan was abandoned due to a lack of suitable space. Even if more GPUs are acquired, storage issues arise due to insufficient available space, power supply, and cooling capacity. The storage limit is 249 GPUs at the first Defense Integrated Data Center in Yongin and 480 GPUs at the second center in Gyeryong.

The military cannot outsource GPU management to private cloud service providers such as Naver Cloud or Kakao. According to the National Intelligence Service’s security guidelines, military and defense sectors are prohibited from using such services. Ultimately, without constructing new data centers, there will be no way to store GPUs, making any plan for AI utilization effectively useless. Due to the nature of national security, new military-owned data centers must be designed as underground facilities with protection against chemical, biological, and radiological threats, as well as electromagnetic pulse (EMP) shielding. The military estimates that the budget required for this will exceed 2 trillion won.

Both inside and outside the military, there are calls to build centers capable of housing large numbers of GPUs, similar to the Ministry of Science and ICT (MSIT). The MSIT plans to invest 2 trillion won by 2028 to establish a “National AI Computing Center” with a capacity of 50,000 GPUs. Additionally, on April 27, it was agreed to actively pursue AI cooperation after a meeting with Demis Hassabis, co-founder and CEO of Google DeepMind.

A military official stated, “We are promoting the construction of additional centers based on the results of research services for data centers, and we plan to conduct AI research through the Defense Artificial Intelligence Technology Research Institute within the Agency for Defense Development.”

This content was produced with the assistance of AI translation services.