CPU ETFs top domestic gains this month as AI agent boom shifts semiconductor investment

The AI semiconductor investment craze is spreading beyond graphics processing units to central processing units. The early commercial success of Meta's personal AI agent Muse has fueled expectations that the proliferation of AI agents will drive a surge in CPU demand. CPU semiconductor ETFs have swept the top spots among domestic ETF gainers this month.

According to Korea Exchange data, the KODEX US CPU Semiconductor TOP10 ETF rose 25.66 percent from Sept. 1 through Tuesday, making it the top-performing domestically listed ETF over that period. The KIWOOM US CPU Semiconductor TOP4+ ETF gained 21.10 percent over the same stretch, ranking second. Both products invest in global CPU companies including Intel, ARM and AMD.

The strong performance was driven in part by the success of Muse, Meta's personal AI agent launched Sept. 8. Muse carries out tasks across multiple services on behalf of users, drawing on their social content from Facebook Marketplace, Instagram and other platforms to generate responses. It also connects through a browser to external services and websites including Gmail, Apple and Google calendars, and OpenTable.

As of Monday, Muse ranked No. 1 in the free apps category on both Apple's App Store and Google Play Store in the United States. The growing expectations around AI agent adoption sent CPU-related stocks sharply higher.

According to Investing.com, Intel rose 27.44 percent over the week from Sept. 14 through Tuesday. AMD gained 26.42 percent, ARM surged 39.41 percent and Qualcomm advanced 10.06 percent.

AMD crossed the $1 trillion market capitalization threshold for the first time on Monday, becoming the fourth chipmaker to join the "trillion-dollar club" after Nvidia, Broadcom and Micron. Kim Dong-won, head of research at KB Securities, said the milestone "signals not merely a rise in the value of individual companies, but a rapid expansion of the AI infrastructure investment axis from a GPU-centered focus to encompass CPUs and memory chips as well."

As AI evolves from chatbot-style systems to agent-based ones, the importance of CPUs is growing. Chatbot AI was limited to answering user queries, but AI agents must carry out multi-step reasoning and tool control sequentially. This increases CPU usage, as the chip coordinates complex task workflows.

The CPU interprets user requests, determines the order of operations and calls up various tools to orchestrate the entire workflow — a role distinct from that of the GPU, which specializes in parallel computation for processing large volumes of data simultaneously.

In a report released Tuesday, global investment bank JPMorgan said CPUs for agentic AI carry 30 to 100 percent more memory than standard CPUs. The bank also projected that the number of GPUs per CPU in AI head nodes would fall from four in 2024 to 1.9 in 2028, signaling a rising share of CPUs relative to GPUs in AI infrastructure.

Shin Seung-yun, a researcher at Hanwha Investment & Securities, said traditional chatbots were overwhelmingly dependent on GPUs for large-scale matrix operations centered on one-off question-and-answer exchanges. "When the shift to AI agents occurs, sequential task processing — including virtual environment and tool control as well as continuous reasoning — becomes central, which will significantly expand CPU usage," he said.

Some analysts warn that CPU supply shortages could worsen. Park Jae-hwan, a researcher at Eugene Investment & Securities, said Muse's success "suggests the growth potential of consumer-facing AI agent services," adding that "the faster agent services proliferate, the more likely CPU demand for workload orchestration will structurally expand — and the already tight CPU supply bottleneck is likely to tighten further."