Meta's newly launched Muse personal AI agent is on track to create an unprecedented hardware bottleneck, requiring up to 1.58 million AMD EPYC CPUs to support just 100 million users. Because Meta promises each user a dedicated cloud-based Virtual Machine (VM) that runs continuously in the background, the sheer scale of compute, memory, and storage needed is staggering. The agent is designed to execute multi-step tasks, such as navigating websites and managing chores, even after the user closes the application.
To ensure privacy and security, Muse operates inside an isolated cloud computer equipped with its own browser to house user data and credentials. Each user's VM is allocated 2 vCPUs, 8GB of RAM, and 100GB of SSD storage. A secondary guardrail system, dubbed Sentinel, continuously monitors the environment to prevent the agent from finalizing any transactions without explicit user confirmation.
Every Muse user is supposed to get their own cloud PC. 2 vCPUs, 8GB RAM, 100GB disk. If Meta actually leaves those boxes on, user growth turns into a chip and memory problem.
- dylan ツ, X
The Staggering Cost of 100 Million Cloud PCs
If Muse reaches 100 million users - a highly plausible target given its current position on app download leaderboards - the infrastructure demands become astronomical. Providing 8GB of RAM and 100GB of storage per user translates to 800 petabytes of RAM and 10,000 petabytes (10 exabytes) of SSD storage globally. While not every user will consume their entire storage allocation, the persistent 24/7 nature of the VMs means the baseline memory requirement remains rigidly fixed.
On the processing side, assuming 126 cores per AMD EPYC processor, Meta would need 1.58 million CPUs for a 100 percent active user base. Even with aggressive resource sharing, the numbers remain massive. If only 10 percent of the user base is actively running tasks at any given time, the load drops to 158,000 CPUs. A more realistic 50 percent network scalability target would still require 790,000 AMD EPYC units just to keep the service stable.
Preparing for the Compute Crisis
Meta is currently offering up to 100 million free tokens per week, with paid subscriptions for heavy power users starting at around $20 per month. To prepare for this massive influx of compute demand, the company has been aggressively expanding its hardware portfolio. In March 2026, Meta became the primary co-developer and lead deployment partner for the Arm AGI CPU.
By April 2026, the tech giant had also integrated tens of millions of AWS Graviton cores into its infrastructure. These strategic acquisitions highlight Meta's awareness of the impending silicon bottleneck. If the company eventually integrates Muse access directly into WhatsApp and Instagram, the current hardware projections will scale exponentially.
The Silicon Ceiling for Personal AI
The hardware math behind Meta Muse reveals a fundamental shift in how AI companies must operate: we are moving from shared inference models to dedicated, persistent cloud computing for every individual. Offering a 24/7 Virtual Machine with 100GB of storage and dedicated vCPUs is a logistical nightmare that mirrors the early, cash-burning days of cloud storage wars, but at an exponentially higher cost per user.
At a $20 monthly subscription tier, Meta is likely operating Muse as a massive loss leader. The raw electricity, cooling, and depreciation costs of maintaining 1.58 million AMD EPYC CPUs and 10 exabytes of SSD storage will quickly outpace subscription revenue. If Meta successfully scales this to its broader social media ecosystem, it won't just dominate the personal AI market - it will effectively corner the global supply of enterprise-grade silicon and memory, forcing competitors into a hardware arms race they cannot afford.
