Central processing units are emerging as a major beneficiary of the artificial intelligence boom as services shift from chatbots to AI agents that perform tasks. Meta’s personal AI agent Muse, launched on the 8th of this month, has helped focus attention on that transition after reaching 1.8 million cumulative downloads in 12 days and 640,000 daily active users, according to Eugene Investment & Securities. In South Korea, KODEX US CPU Semiconductor TOP10 rose 22.65% from Sept. 16 to 23, the strongest gain among domestic ETFs tracked by ETF CHECK, while ACE Global AI Custom Semiconductor gained 19.39% and KIWOOM US CPU Semiconductor TOP4+ rose 16.18%. KODEX’s largest holdings as of the 23rd were Intel at 26.59%, AMD at 25.96% and Arm at 24.94%, together representing more than 77% of the fund. AI agents repeatedly access databases, call external programs and run new inference cycles, increasing the coordination workload handled by CPUs even as GPUs perform much of the core AI computation. Muse assigns a dedicated cloud virtual machine to every user, including free-tier users. Intel has projected that the GPU-to-CPU ratio could move from between eight-to-one and four-to-one currently to one-to-one or higher over the long term, while AMD has estimated the server CPU market could reach $220 billion by 2030. The potential increase in server CPU deployments is also supporting expectations for demand for multilayer ceramic capacitors, Ajinomoto build-up film semiconductor package substrates and high-end printed circuit boards. Meritz Securities said Samsung Electro-Mechanics signed a long-term MLCC supply contract with a North American CPU maker and raised its target price to 2.2 million won from 2.1 million won. Shinhan Securities cited demand for advanced PCBs at Isu Petasys and raised its target price to 170,000 won from the previous level, while estimating the company’s second-quarter PCB average selling price increased 21% from a year earlier. Eugene Investment & Securities analyst Park Jae-hwan said faster adoption of agent services could structurally expand demand for CPUs handling workload orchestration and further intensify the existing CPU supply bottleneck.