The explosive popularity of an AI assistant has ignited global technology stocks. In September, Meta launched its AI assistant Muse, which climbed to the top of the US App Store free charts in less than two weeks. On September 21st, Meta surged 11% in a single day, AMD's market capitalization broke the $1 trillion mark for the first time, and both Intel and ARM rallied sharply. The following day, CPU-related stocks on the A-share market moved higher in tandem. AI is no longer just about conversation; it is starting to perform real tasks. The foundational logic of the computing power world is being fundamentally rewritten.

The Consumer-Facing Application Narrative Gains Traction: What Exactly is Muse?

On September 8th, Meta officially launched its personal AI agent, Muse. Within less than two weeks of its release, it directly topped the US free app charts on both Apple and Google stores, surpassing 2.5 million downloads and reaching 642,000 daily active users—nearly three times the figure ChatGPT recorded in its equivalent early period. The capital markets were instantly ignited. Meta's stock price surged 11.43% in a single day, its biggest gain since April 2025; AMD jumped nearly 10%, with its market value breaking the $1 trillion threshold for the first time; Intel climbed over 12%, and ARM skyrocketed 17%. The A-share market followed suit, with CPU concepts collectively strengthening and leaders like Hygon Information Technology performing well. So what exactly is Muse? Simply put, it is not just a chatbot; it is an AI assistant capable of taking action on behalf of users. Meta equips each user with a dedicated virtual computer in the cloud, capable of connecting to services like Gmail, calendars, and OpenTable, helping users fill out forms, manage emails, shop, book travel, and even continue working in the background after the app is closed. Muse represents a tangible signal that Meta's hundreds of billions of dollars in AI investment are beginning to pay off. Advertising, subscriptions, and transaction revenue sharing could all become new monetization avenues. This also means the narrative for consumer-facing AI applications is now validated.

Shifting Compute Dynamics: Is the CPU Set to Play the Leading Role?

Muse's significance extends far beyond being just another product. It triggers a repricing of the entire underlying logic of AI compute infrastructure—as AI transitions from "answering questions" to "executing tasks," the demand structure for computing power is being rewritten. Logic One: AI Moves from "Talking" to "Doing," Changing Compute Demand. In recent years, the main thread of AI investment has revolved around GPUs—large model training requires massive parallel computing, an architecture for which GPUs are naturally suited. However, the operational model of an Agent is entirely different. An Agent can be broken down into an "LLM (Brain) + Planning (Orchestration) + Tool Use (Execution) + Memory" framework, where the CPU handles critical functions such as task orchestration, data scheduling, sandbox execution, memory management, and tool invocation. The large-scale deployment of Agentic AI is "completely restructuring CPU compute demand," breaking the traditional AI paradigm where the CPU plays a supporting role and the GPU leads. Further analysis shows that AI workloads are shifting from compute-intensive to system-intensive. Task scheduling, tool invocation, sandbox operations, and data parsing can all run on CPUs, significantly elevating the CPU's importance within AI systems. This means the CPU's role is upgrading from a "data porter" to a task-leading "chief of staff," with its workload shifting from pulsed to sustained full-load operation. The surge in concurrency and isolation requirements is driving CPU demand to grow at a "multiplicative" rate. Logic Two: Not a Slight Increase, but a Fourfold Jump—CPU Demand Has Completely Exploded. The growth in CPU demand during the Agent era is not a simple linear increase but a leap in magnitude. ARM projects that a traditional AI data center requires approximately 30 million CPU cores per gigawatt. In the Agentic AI era, this number is expected to skyrocket to 120 million cores—a full fourfold increase. ARM CEO Rene Haas further pointed out that the industry is likely to shift towards CPU designs with 256 or even 512 cores. At the allocation level, the CPU-to-GPU ratio is also undergoing fundamental change. AMD CEO Lisa Su noted that the ratio of CPUs to GPUs in AI infrastructure deployment is shifting from the past 1:4 or 1:8 towards nearly 1:1, and in scenarios with a massive increase in the number of agents, CPU counts may even surpass GPUs. CITIC Securities research suggests that in an optimistic scenario, the CPU-to-GPU ratio could reach 8:1, with the long-term market size potentially reaching $2 trillion, surpassing the current GPU market. Morgan Stanley estimates that driven by surging demand for AI agent infrastructure, the potential CPU total addressable market (TAM) could reach $125 billion by 2030. AMD projects a compound annual growth rate of over 35% for the total addressable server CPU market in the coming years and has doubled its 2030 market size forecast from last year's $60 billion projection to $120 billion. Logic Three: Orders Booked Out Six Months Ahead—CPU Market Enters "Panic Buying" Mode. While demand explodes, the supply side is also under strain. Intel CEO Lip-Bu Tan has stated that the company can currently only fulfill about 50% of its CPU customer orders. Furthermore, Intel plans to raise PC CPU prices by approximately 10% again in October. This supply-demand imbalance is creating a scenario of both rising volumes and higher prices. Agents bring a massive, order-of-magnitude increase in non-human callers, where every task orchestration, retrieval, and tool invocation heavily depends on CPU compute power. The implementation of AI inference and Agent applications is expected to continue driving server CPU demand into a cycle of supply shortage, rising volumes, and increasing prices. Which Beneficiaries Should Investors Focus On? Following this logic chain of "Agent Explosion—CPU Demand Surge—Supporting Upgrades," three main investment themes emerge from an industry perspective. First, the CPU-related segment. The scaled operation of Agents pushes the CPU from a supporting role to the core of task scheduling. Server CPU demand is shifting from linear growth to a magnitude leap. Overseas CPU manufacturers benefit from supply constraints and rising prices, while domestic CPU makers benefit from the advancement of Xinchuang (domestic IT infrastructure) and accelerated self-reliance, with penetration rates expected to rise in key sectors like government, finance, and telecommunications. Second, overseas compute expansion represented by Meta. Cloud vendor capital expenditure continues to increase, and data center cluster sizes are expanding. As a critical link for compute interconnection, optical modules show strong demand certainty. Rapid iteration of high-speed products and longer order visibility support high industry prosperity, upgrading the supply chain from supporting components to a core track within the compute infrastructure. Third, the application segment. Muse is driving the penetration of agents from the developer community to general consumers. The 2C Agent is poised to become a significant direction for AI application implementation. Software and platform companies with advantages in scenarios, data, and distribution are poised for value reassessment, and the commercial potential of the AI application layer is gradually opening up.

Relevant ETFs to Watch

ChiNext Artificial Intelligence ETF Huabao (159363): Combining optical modules and AI applications, this ETF offers a dual-driver approach of ChiNext compute power and applications. The underlying index includes over 35% weight in "Zhongji Innolight + Eoptolink + TFC Communication," positioning it at the forefront of optical technology with a focus on leading CPO optical module companies, serving as a core flag bearer for AI compute. Feeder funds: Class A 023407, Class C 023408.

STAR Chip ETF Huabao (589190): Featuring a relatively lower fee rate among its peers, this ETF covers the entire chip industry chain, with heavy positions in AI chips, memory, and semiconductor equipment and materials. It effectively maps the industry development trend driven by the continuous upgrade of AI compute infrastructure and offers strong offensive characteristics. Feeder funds: 021225.

STAR Chip Design ETF Huabao (589430): Adopting a Fabless asset-light model, this ETF focuses on the high-elasticity pure design segment. The underlying index covers four main areas: compute, storage, interconnect and custom, and featured perception and SoC. It holds the core strength of Chinese chip design and is a key standard-bearer for AI compute and domestic substitution.

Hong Kong Internet ETF Huabao (513770): This is the core ETF for Hong Kong-listed AI applications, gathering Hong Kong's internet giants. The underlying index heavily weights major cloud vendors like Alibaba-W and Tencent Holdings, along with various AI application companies. It boasts significant leadership advantages and supports intraday T+0 trading. Feeder funds: Class A 017125, Class C 017126.

Note: Please refer to the respective fund legal documents for fee details. Reminder: Recent market volatility may be significant, and short-term gains or losses do not predict future performance. Investors are advised to invest rationally based on their own capital situation and risk tolerance, with close attention to position sizing and risk management. Data sources: Shanghai and Shenzhen Stock Exchanges, Wind, etc. Note 1: As of 2026.9.21, according to Guozheng Index, the top three constituent stocks of the ChiNext Artificial Intelligence Index are Eoptolink (13.94%), Zhongji Innolight (12.87%), and TFC Communication (10.57%). Note 2: The STAR Chip ETF Huabao has a management fee of 0.3%, custody fee of 0.08%, and a total fee rate of 0.38%, which is relatively low among ETFs tracking the same underlying index. Note 3: As of 2026.9.21, according to the CSI Index, the top two constituent stocks of the CSI Hong Kong Stock Connect Internet Index are Alibaba-W (14.22%) and Tencent Holdings (13%). Based on the fund manager's assessment, ChiNext Artificial Intelligence ETF Huabao, STAR Chip ETF Huabao, STAR Chip Design ETF Huabao, and Hong Kong Internet ETF Huabao carry a risk rating of R4 (medium-high risk), suitable for aggressive investors (C4) and above. Please refer to the sales institution for suitability matching opinions. Risk Disclosure: The STAR Chip Design ETF Huabao passively tracks the SSE STAR Market Chip Design Thematic Index, with a base date of 2019.12.31 and a release date of 2024.7.26; the Hong Kong Internet ETF Huabao passively tracks the CSI Hong Kong Stock Connect Internet Index, with a base date of 2016.12.30. The index constituent composition is adjusted according to the index compilation rules, and backtested historical performance does not indicate future index returns. The constituent stocks mentioned in this article are for display purposes only, and descriptions of individual stocks do not constitute investment advice in any form, nor do they represent the holdings or trading activity of any fund under the manager. Any information appearing in this article (including but not limited to individual stocks, comments, forecasts, charts, indicators, theories, and any form of expression) is for reference only. Investors are solely responsible for their own independent investment decisions. Furthermore, any views, analyses, and forecasts in this article do not constitute investment advice to readers in any form, and no liability is assumed for any direct or indirect losses resulting from the use of the content herein. Fund investment carries risks. Past performance of a fund does not represent its future performance. The performance of other funds managed by the fund manager does not constitute a guarantee of a fund's performance. Please invest in funds with caution. A MACD golden cross signal has formed, and these stocks are showing good upward momentum!