As Lisa Su, Chairman and CEO of AMD, is expected to visit Korea next month, attention is turning to the agenda for cooperation between Korea and AMD in the artificial intelligence (AI) industry. The Korean government and AMD previously signed a memorandum of understanding (MOU) on cooperation in the AI semiconductor ecosystem, and the key issue will be how specifically they will define collaboration tasks centered on neural processing units (NPUs).
According to government and industry sources on the 30th, an industry roundtable attended by Chairman Lisa Su is being discussed for the 8th of next month in Seoul. Chairman Lisa Su visited Korea in March this year, holding a series of meetings with CEOs from the domestic semiconductor and IT industries, including Lee Jae-yong, Chairman of Samsung Electronics, to discuss cooperation methods for each company. This visit to Korea is also expected to be an extension of the project to strengthen cooperation with Korea. The Ministry of Science and ICT is reportedly coordinating a meeting with Deputy Prime Minister Baek Keun-hoon during this visit schedule.
Previously, Deputy Prime Minister and Minister of the Ministry of Science and ICT Baek Keun-hoon met with Chairman Lisa Su in San Francisco, USA, last July to sign an MOU for cooperation in the fields of AI semiconductors and computing. The plan is to build an AI ecosystem by partnering with AMD, one of the two major global CPU companies.
Specifically, the core is to build an open AI infrastructure combining CPUs and GPUs produced by AMD with domestically made NPUs. Through collaboration between major domestic NPU companies such as Rebellions, DeepX, and Mobileye, and AMD, they aim to build heterogeneous AI computing infrastructure and proceed with demonstration projects. The plan is to diversify AI computing resources while enabling the use of domestic NPU products for AI computation.
The government's move to create an AI ecosystem through NPUs stems from the judgment that if the current GPU-centric AI ecosystem continues, the risk burden associated with rising GPU prices will inevitably increase.
For example, in the 2027 budget plan announced earlier this month, the Ministry of Science and ICT included 940 trillion won for AI projects, of which 423.85 billion won was allocated specifically for securing GPUs. While a "generous" budget allocation was made, including 27.5 billion won per unit for the next-generation chip 'Vera Rubin,' it is predicted that the quantity secured could decrease due to rising GPU prices even with this amount.
The fact that different AI services have varying requirements for training, inference, and data processing, making NPU-centric computation advantageous over GPUs in certain fields, is also cited as a reason why NPU-centric collaboration is necessary. Plans for domestic NPU companies to cooperate with AMD to be incorporated into the global AI semiconductor supply chain, including servers and data centers, are also expected to be seriously considered.
A government official explained regarding cooperation with AMD, "It would be a considerable burden even for Korea, as a follower, if the AI ecosystem were built solely around GPUs." The official added, "Since some AI services run better on NPUs than on GPUs, we are trying to propose alternatives like NPUs to the market so that diverse choices can be made in the process of creating an AI ecosystem."
In addition, the government plans to focus on cultivating domestic talent by establishing excellent AI research centers in Korea through cooperation with AMD. The goal is to create a collaboration hub that supports technology cooperation among domestic companies, universities, and research institutions, AI semiconductor software development, computing technology verification, and joint research, based on AMD's AI computing resources and software capabilities. The plan is to expand cooperation with domestic universities and research institutions by building joint testbeds utilizing domestic NPUs through the AI research center, and to carry out research projects for creating an open AI ecosystem.
in TECH
