A claim has emerged that the government should become the first customer for graphics processing units (GPUs) to be secured by 2030 and lead the early market.
The AI Policy Cooperation Committee under the Korea Artificial Intelligence and Software Industry Association (KOSA) stated accordingly through the Public sector GPU utilization strategy report published on the 19th. The report warned, "Given that the lifespan of GPUs is short at three to five years and that there is a lack of practical demand relative to the scale of infrastructure, if utilization is not maximized from the introduction stage, an asset into which an enormous budget has been invested risks devolving into 'scrap metal.'"
It added, "We must boldly shift the center of gravity of national policy from the existing competition to possess infrastructure to a competition to utilize it in industrial fields."
In the report, the AI Policy Cooperation Committee stressed that the government should become the "first customer (First Customer)" and lead the early market. The committee said, "By mandating the adoption of domestic AI in administration and defense in the public sector, we should remove market uncertainty, and for small and midsize manufacturing companies with low AI adoption rates, it is necessary to establish a 'one-stop package' that supports the entire process from diagnosis to implementation."
It also advised changing the budget structure, which is skewed toward hardware purchases, to recognize the value of software (SW) and data, and strategically separating the training and inference stages to improve infrastructure efficiency. The report said, "For high-difficulty model development (training), concentrate the deployment of Nvidia GPUs, but in the stage of public-facing services (inference), make the use of domestic Neural Processing Unit (NPU) chips the rule to support securing early references for domestic chips."
Lastly, it added that the country needs to cultivate practical AI engineering talent and "supercomputing architects." It explained that, alongside up-skilling to equip industry veterans with domain knowledge with AI capabilities, the nation should systematically train architects who can optimally design and operate large-scale GPU clusters.
Im Woo-hyung, Chairperson of the AI Policy Cooperation Committee (co-head of LG AI Research), said, "Now that GPU acquisition is becoming tangible is the golden time to realize a leap to the three leading AI nations (G3)," adding, "A 'one-team' play between the public and private sectors, in which the public sector plays a priming role preemptively and the private sector responds with creative engineering, is more urgent than ever."
Cho Joon-hee, president of KOSA, said, "Korea must build a 'full-stack AI' package using the world-class 'manufacturing data' it possesses as a weapon to secure global export competitiveness," adding, "The association will take the lead so that the government budget does not remain only in hardware such as GPU purchases, but leads to the creation of a healthy ecosystem in which software value is properly recognized, such as paying fair AI process fees."