The AI semiconductor market is rapidly growing. With NVIDIA accounting for 90% of global sales, voices are rising that domestic fabless companies should seize opportunities in the expanding AI semiconductor market.

On the 25th, a forum was held at the National Assembly Members' Hall in Yeouido to discuss ways to foster an AI semiconductor ecosystem and strengthen global competitiveness. This National Assembly debate, titled "Korea in the Age of Hyper-Giant AI and the AI Semiconductor War," was attended by three promising domestic fabless companies in the AI semiconductor sector.

On this day, Sapion Korea CEO Ryu Su-jeong, Furiosa AI CEO Baek Jun-ho, and Rebellion CEO Park Sung-hyun attended the discussion and expressed their opinions on Korea's AI semiconductor development plan and government support policies.

Park Sung-hyun, CEO of Rebellion, said, “In the past, when Korea raised its voice for DRAM development, domestic top-tier experts actually voiced opposition,” and added, “When everyone else was shouting that it couldn’t be done, the government’s policy momentum served as a springboard for semiconductors to become Korea’s largest foreign currency-earning industry.”

Park, who said that the domestic semiconductor industry, which had been at the global top of DRAM and NAND flash even in more difficult situations, is in a better position in the AI semiconductor competition compared to then, added, “The market is expanding after Chat GPT, the size of the new market is very large, and there are large companies in Korea that will consume AI semiconductors, such as Naver and Kakao.”

Baek Jun-ho, CEO of Furiosa AI, also emphasized the excellence of Korea's semiconductor base. Citing Samsung's established presence in the development of smartphone modem chips, its world-leading technology in SSDs, and its unique foundry capacity, including TSMC, the report evaluated Korea, China, and the United States as the only countries capable of producing finished products.

Ryu Su-jeong, CEO of Sapion Korea, said, “Cloud service companies are developing their own AI accelerators, but the market continues to grow to the point where even they cannot respond to all the various application groups,” adding, “It is not easy to make semiconductors that can keep up with the rapidly changing service trends.”

The AI semiconductor development cycle is short, lasting one to two years. Therefore, a gap is inevitable between the pace of service development and the pace of hardware support. This gap is expected to provide opportunities for domestic fabless companies.

Ha Jung-woo, head of Naver AI Lab, who attended the discussion that day, expressed his opinion as a company in demand for AI semiconductors, saying, “Super-large AI will lead to various services and products in daily life,” and advised, “There are differences between AI semiconductor design and service characteristics, so we need to release semiconductor products needed in the field that reflect the technological needs of service companies into the market.”

AI semiconductors are expected to expand the market through specialized and customized strategies by solving cost issues in AI construction and operation for demand companies along with low-power advantages and developing semiconductors that deliver optimal performance suitable for various service groups.