On July 16th, at the “2026 Korea Businessmen's Association (FKI) CEO Jeju Summer Forum,” June Paik, CEO of AI chip design company FuriosaAI, stated, “It is projected that by 2030, the global scale of newly built AI data centers will reach 100 gigawatts. In terms of semiconductor procurement demand, this means a market worth tens of trillions of Korean won is about to form.” As the focus of generative AI development gradually shifts from “training” for building models to “inference” supporting practical applications, industry insiders predict that the market size for inference chips, which combine high energy efficiency and low power consumption, is expected to reach tens of trillions of Korean won in the future.
June Paik predicts that most newly built AI data centers in the future will primarily deploy inference computing power, rather than training computing power. He pointed out that the decisive factor in the competition for inference AI chips is not computing power, but energy efficiency*. Since power supply has become the biggest bottleneck in building AI data centers, chips that can complete more AI computing tasks with the same power consumption will be more favored by the market. He also likened the development of AI chips to the transformation of the automotive industry from gasoline-powered cars to electric cars. He stated that GPUs still dominate the AI computing market, but in order to further improve energy efficiency, more dedicated inference chips optimized for specific AI computing tasks will be needed in the future.
