Meta’s Muse AI agent puts server CPUs back in the spotlight

Meta's Muse AI Agent Puts Server CPUs Back in Spotlight

Introduction

For the past few years, the AI hardware story had one main character: the GPU. However, Meta's recent launch of its Muse AI agent on AMD EPYC processors has handed a speaking role to the humble server CPU.

What Muse Does

Muse is designed to handle multi-step tasks on its own, such as booking travel, sending emails, and managing projects through apps connected to it.

Technical Details

Under the hood, Meta built the system on AMD EPYC Turin processors, specifically the EPYC 9D25 models, which feature up to 128 cores. Muse runs on the EPYC 9D25, according to an earnings report.

User Adoption

Each Muse user gets an Ubuntu sandbox with 2 virtual CPUs and 8 GB of memory. Users showed up, and so did investors, with Muse topping the US app store rankings shortly after launch and passing 500,000 daily active users in that early window.

Market Impact

Estimates of CPU-to-GPU resource ratios for agentic workloads now range from 4:1 to 40:1 in specific scenarios. Market forecasts suggest the server CPU market may surpass $120 billion by 2030 as businesses adjust to the compute demands of agentic AI.

Conclusion

For AMD, Muse is the clearest public validation yet of its server processor strategy in the agent era. Being the named hardware behind both Meta's and reportedly OpenAI's personal agents gives it a strong narrative heading into the next round of data center spending decisions.