The rise of autonomous AI agents is changing the kind of silicon that matters. Meta's Muse, OpenAI's Dots, and coding tools like Codex use AI models to plan tasks that run on their own. These agents need a computer that can run for hours or days in the background, and that means CPUs are doing the workflow while GPUs do the thinking, according to Futurum Group CEO Daniel Newman.
Meta and OpenAI run agents on virtual machines that partition servers into lighter virtual computers. AMD's EPYC processors, with up to 192 cores, fit this role. Benchmarks show Muse's virtual computer uses two cores and Dots uses nine. Dots runs on an AMD EPYC 9V74, available for under $3,000, while the EPYC 9D25 reportedly used by Meta costs even less on the secondary market. By contrast, Nvidia GPUs can cost more than 10 times that per chip and are typically sold in clusters of hundreds or thousands.
Investors are watching how Meta will serve Muse users, especially after downloads passed 5 million since last month's launch, per Sensor Tower. Morgan Stanley estimates Muse's serving costs at $3 to $130 per month, averaging $37 per user depending on inference usage. The firm says Meta's agent could account for 20% of AMD's 2026 chip sales.