Meta's artificial intelligence agent Muse is pushing CPUs from a supporting role in the AI narrative into the spotlight. The latest report from TrendForce shows that CPU delivery lead times have extended from the usual 16-20 weeks to 25-30 weeks, and it directly attributes this change to the pull on CPU demand from Muse and other AI agents.

This marks the first time the market has obtained direct supply-chain evidence of AI agents impacting CPUs. Previously, discussion of Muse mostly revolved around user growth and conceptual speculation. Now, the significant lengthening of lead times means the squeeze that AI agents place on hardware infrastructure has spread from GPUs to CPUs.

Yet there is still a clear gap between expectations and actual demand scale. The 12.5 million cores estimated by Altimeter partner Freda Duan is closer to a theoretical upper limit, and Meta may not strictly deliver one dedicated virtual machine for every user. There is still a verification distance between bullish expectations and realized orders.

Longer lead times, but the bottleneck remains unresolved

TrendForce's report provides a clear incremental data point: CPU lead times have extended from 16-20 weeks to 25-30 weeks, and it lists Muse and other AI agents as a core factor driving demand higher. Over the past two years, discussions of AI infrastructure have been almost entirely centered on GPUs. Now, the pressure AI agents exert on hardware has been transmitted to CPUs for the first time.

However, the report does not specify exactly where the delivery extension is stuck. The CPU supply chain involves three major stages: wafer foundry, packaging, and testing. Bottlenecks in different stages correspond to very different supply-demand dynamics and alleviation timelines. At present, CPU manufacturers, server OEMs, and cloud providers have not publicly confirmed where the problem lies, and this information gap itself is an important market variable.

12.5 million cores: a theoretical ceiling, not actual orders

Altimeter partner Freda Duan's estimate provides the market with a quantifiable anchor for discussion. She assumes Muse reaches 100 million daily active users and that each user averages 2 hours of use, which would require about 8.33 million virtual machines to meet average load. When peak load is 2.5 times the average, the maximum number of virtual machines would reach 20.8 million. Combining that with the physical cores required per virtual machine, using DeepSeek as a reference but assuming Meta's operational efficiency is lower than DeepSeek's, she concludes that under an assumption of 25 million active virtual machines, Meta may need about 12.5 million CPU cores.

Duan herself is cautious about this figure. She points out that AI agent sandboxes and virtual machines generally suffer from resource over-allocation, and when virtual machines remain running, actual CPU utilization is usually close to zero. Therefore, Meta does not need to strictly deliver the promised hardware for every user, and 12.5 million cores is closer to a theoretical upper limit than actual procurement demand. She also cautions that as inference speed improves, the number of physical cores allocated per virtual machine will further decline. But higher technical efficiency is often accompanied by demand expansion, so total CPU demand may not necessarily slow as a result.

Intel and AMD's unexpected variable

Meta has promised to provide every Muse user with a dedicated virtual machine equipped with 2 virtual CPUs, 8GB of memory, and 100GB of solid-state storage. Within ten days of Muse's launch, its user base reached 700,000.

Preliminary calculations show that as Muse's daily active user scale expands, it could spur demand for millions of additional CPUs. The market therefore believes CPU manufacturers have already received a structural boost, but this expectation has not yet been reflected in specific revenue guidance or order data.

Lead-time data is the first hard signal. What needs attention next is how Intel and AMD describe their data center CPU businesses in next quarter's earnings reports, as well as changes in the CPU-to-GPU ratio within cloud providers' capital expenditures.

Whether longer lead times can translate into realized revenue on the financial statements is the key verification point for this round of the CPU narrative.