
SILICON VALLEY — Meta's artificial intelligence agent Muse has taken off, driving up the value of central processing unit makers. Demand for CPUs, which boost inference capability, is expected to grow further as AI agents spread from businesses to individual users. Industry watchers said Muse succeeded because, unlike other personal AI assistants, it focuses on everyday tasks and keeps the number of steps to a minimum.
Bloomberg reported on the 21st, citing data from research firm Sensor Tower, that Muse ranked first in free app downloads that day on both Apple's App Store and Google Play in the United States. That put it ahead of OpenAI's ChatGPT just 13 days after its launch on Sept. 8. Meta shares closed at $741.24 in New York, up 11.43% from the previous session. The stock has gained more than 11% so far this year.
Meta has bet on a personal, everyday-task agent to take on Anthropic and OpenAI, which dominate the corporate AI agent market. The agent drafts and sends emails on a user's behalf, makes travel bookings and helps with online shopping. Google Gemini's Spark and the agent mode in OpenAI's ChatGPT had already billed themselves as personal AI assistants before Muse, but reviewers say Muse is more convenient for handling everyday chores.

Muse's strength lies in reading a user's context across not only Meta's own social media services, such as Facebook Marketplace and Instagram, but also outside platforms including Google's Gmail. Chatbots require users to issue commands at every step, while fully automated AI agents are convenient but carry the risk of malfunction. Muse, by contrast, works on its own with minimal further instruction once a user sets a goal. It leaves the final send or payment step to the user, reducing the chance of errors. Users have praised it as "an AI that gets the work done while I'm not paying attention and leaves only the last decision to me."
Its business model is also seen as effective in widening its user base. Subscribers get up to 100 million tokens a week and a virtual machine for the agent free of charge, while heavy users pay $20 or $100 a month, and the service takes a commission on shopping transactions.
Meta's strategy has paid off. The company had been criticized for pouring $200 billion into AI without producing a credible model, but Muse has turned the mood around. "Most of the agent use cases so far have not been for ordinary people," Bernstein analyst Stacy Rasgon said in a CNBC interview, adding that a number of agents including Meta's Muse are now starting to show what is possible.
Muse's success is good news for CPU makers, because CPUs determine the inference capability at the heart of AI agents. Graphics processing units handle AI training and computation in data centers, while CPUs coordinate the GPUs' processing to produce the results users want. If GPUs are the players, CPUs are the coach. TrendForce estimates that as AI agents spread, the ratio of CPUs to GPUs in data centers could shift from 1-to-8 to 1-to-1.
AMD and Intel are the leading makers of CPUs for AI servers. AMD closed at $615.52, up 9.95% from the previous session, becoming the fourth U.S. chipmaker to pass $1 trillion in market capitalization (about 1,354.3 trillion won), after Nvidia, Broadcom and Micron. Intel jumped 12.14%.
Arm Holdings surged 17.16%. Arm, the world's largest semiconductor intellectual property company, announced its entry into the server CPU market in March this year with the release of its first in-house chip. Qualcomm rose 9.29%. The company unveiled its Dragonfly C1000 server CPU in June and said it would supply the chip to Meta starting in 2028. The rally in chip stocks pushed the tech-heavy Nasdaq composite to a record close.







