Central China: Muse sparks a surge in personal agent applications; the importance of CPUs continues to rise in the Agentic + AI era.

Central China released a research report stating that recently, multiple personal AI agents such as Muse, Dots, and Instinct have been launched one after another. The autonomous planning, multi-step reasoning, and external tool invocation required by AI agents have shifted workloads from compute-intensive to control-intensive, branch-intensive, and IO-intensive. GPUs handle computation, while CPUs determine whether computation can continue efficiently. In the Agentic AI era, CPUs are returning to the core of AI data centers, and the CPU-to-GPU ratio is shifting from 1:4~1:8 toward nearly 1:1. The bank believes that the semiconductor industry is still in an upcycle, with AI as an important driver of semiconductor industry growth; server CPU demand is currently strong while supply is tight, and a new round of price increases is expected. It is recommended to pay attention to investment opportunities in domestic CPU manufacturers. Central China's main points are as follows: The domestic semiconductor industry performed relatively weakly in September In September 2026, the domestic semiconductor industry (CITIC) fell 8.83%, while the CSI 300 fell 5.78% over the same period. Among them, integrated circuits fell 8.37%, discrete devices fell 8.93%, semiconductor materials fell 7.38%, and semiconductor equipment fell 10.43%; the semiconductor industry (CITIC) rose 33.78% year-to-date in 2026. In September 2026, the Philadelphia Semiconductor Index rose 9.48%, while the Nasdaq 100 rose 3.23% over the same period, and the Philadelphia Semiconductor Index rose 78.29% year-to-date in 2026. Global monthly semiconductor sales continued to grow rapidly year over year, and domestic and overseas cloud vendors continued to increase capital expenditure In August 2026, global semiconductor sales increased 144.3% year over year and 8% month over month; according to the latest forecast from WSTS, global semiconductor sales are expected to reach $1.655 trillion in 2026, up 108% year over year. Downstream demand shows a trend of structural divergence, with demand for AI computing hardware infrastructure remaining strong. In 26Q2, capital expenditure by the four major North American cloud vendors Google, Microsoft, Meta, and Amazon increased 86% year over year and 27% quarter over quarter, and they continued to raise their 2026 capital expenditure plans; in 26Q2, capital expenditure by the three major domestic internet vendors Alibaba, Baidu, and Tencent increased 102% year over year and 92% quarter over quarter. In September 2026, DRAM spot prices showed some divergence month over month, NAND Flash spot prices declined slightly month over month, the DRAM index rose about 2.19% month over month, and the NAND index fell about 0.42% month over month. According to SEAJ data, global semiconductor equipment sales increased 23% year over year in 26Q2, China semiconductor equipment sales increased 5% year over year in 26Q2, and Japan semiconductor equipment sales increased 47.4% year over year and 7.5% month over month in August 2026; SEMI expects global semiconductor equipment sales to grow 23.2% to $165.9 billion in 2026. Global silicon wafer shipments increased 7.4% year over year and 9.1% quarter over quarter in 26Q2. In summary, the bank believes that the semiconductor industry is still in an upcycle, with AI as an important driver of semiconductor industry growth. Investment recommendations Recently, multiple personal AI agents such as Muse, Dots, and Instinct have been launched one after another, and Meta Muse sparked a surge in personal agent applications; just five days after launch, Muse's downloads surpassed ChatGPT and topped the U.S. iOS free chart; according to estimates by Sensor Tower, Muse's downloads exceeded 5 million within 22 days of launch, while ChatGPT's downloads over the same period were 2.3 million. When AI agents require autonomous planning, multi-step reasoning, external tool invocation, database connection, code execution, and continuous interaction, workloads shift from compute-intensive to control-intensive, branch-intensive, and IO-intensive. GPUs handle computation, while CPUs determine whether computation can continue efficiently. In the Agentic AI era, CPUs are returning to the core of AI data centers. At the Advancing AI 2026 conference, AMD CEO Lisa Su stated that the server CPU market size in 2030 has been raised from the previous $120 billion to $220 billion, with a compound growth rate of more than 50% from 2025 to 2030, making CPUs a core incremental driver of AI infrastructure. In traditional AI data centers, the CPU:GPU ratio is 1:4~1:8. Agents are driving the CPU:GPU ratio from 1:4~1:8 toward nearly 1:1, and the status of CPUs continues to rise. AI inference and agent applications are reshaping the computing power structure, and CPUs are returning to the core of AI data centers. At present, server CPU demand is strong while supply is tight, and a new round of price increases is expected. Domestic CPU manufacturers are accelerating the realization of domestic substitution. It is recommended to pay attention to investment opportunities in domestic CPU manufacturers. Risk warnings: downstream demand falling short of expectations, risk of intensified market competition, domestic manufacturers' R&D