The era of CPUs is coming again.

1/ On September 8, Meta launched the personal AI agent Muse. Unlike a simple response-based chatbot, it is a personal agent capable of proactively executing tasks and generating ideas on behalf of users. Its features include sending emails and booking travel, and its persistent structure allows it to continue performing tasks in the background even after the app is closed.2/ The initial market reaction upon launch was overwhelming. Within 12 days, the accumulated downloads reached 1.8 million, surpassing ChatGPT’s record for the same period, and the DAU also reached 640,000, significantly higher than ChatGPT’s at the same point in its release. This has been seen as the first successful case of an AI agent expanding into the consumer market.3/ The business structure is also conducive to increasing usage. Compared to chatbots, AI agents have longer cycles and higher token consumption, making it a cost-heavy service. However, Meta offers up to 100 million tokens free each week even for non-subscribers. In the long term, its goal is to collect commissions from sellers on shopping, travel, and other platforms integrated with Muse.4/ It is worth noting that this workload is CPU-intensive. AI agents involve multi-step loop structures that require external tool calls, DB access, and repeated inference. Aside from AI computing handled by the GPU, decision-making, orchestration, and external software access are all within the CPU's domain.5/ Especially with Muse’s structure, where even free users are assigned their own dedicated cloud VM, it is likely that as users increase, CPU demand will scale proportionally. Intel has noted that as agents proliferate, the GPU:CPU ratio, currently at 8~4:1, may eventually rise to over 1:1.6/ Meta’s investment movements are also pointing in this direction. In the first half of the year, Meta secured broad contracts for major server CPUs such as NVIDIA Grace·Vera, ARM AGI, AMD Venice, and AWS Graviton. This is interpreted as early strategic planning for the expansion of agent services, including Muse.7/ The popularity of Muse demonstrates the growth potential of consumer-oriented AI agent services. The faster agent services spread, the more likely that structural CPU demand for workload orchestration will expand.8/ The already tight CPU supply bottleneck is likely to intensify further. The major CPU manufacturers such as Intel, AMD, ARM and the CPU value chain, which recently went through adjustments, may regain attention. I believe it's time to consider taking a buying position.