{"slug": "meta-s-muse-may-need-2000-server-trays-at-reported-user-scale", "title": "Meta's Muse may need 2,000 server trays at reported user scale", "summary": "Tom's Hardware estimated in a September 25th report that Meta's Muse agent could require roughly 2,000 server trays at its reported scale of more than 500,000 daily active users, based on user-observed sandboxes with two vCPUs and about 8GB of memory and an assumed density of 256 environments per 512-vCPU, 2TB server. The outlet called the figure rough napkin math, and Meta has not published per-user VM specifications, server density or fleet capacity; Meta's September 8th launch announcement describes Muse as running on a dedicated cloud computer. User tests by blogger Evan Hoffman and analyst Tae Kim identified AMD EPYC 9D25 processors, Ubuntu 24.04 and Linux kernel 7.0, with CPU-only sandboxes and separate GPU servers handling inference, while a separate hands-on test measured one Muse instance at 7.7GB of RAM and 100GB of persistent storage.", "body_md": "# Meta's Muse may need 2,000 server trays at reported user scale\n\n**Tom's Hardware's estimate combines Meta's Muse's reported 500,000-plus daily active users with 256 sandboxes per hypothetical tray. The figure is napkin math, not a disclosed Meta deployment.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nPrimary source: [Tom's Hardware](https://www.tomshardware.com/pc-components/cpus/meta-muse-runs-agents-on-amd-epyc-turin-hosts-with-two-cores-and-8gb-of-memory-ai-agent-can-pass-terminal-commands-to-ubuntu-host-system)\n\n## Why it matters\n\nThe reported CPU and memory allocation offers a concrete, if unconfirmed, basis for estimating the infrastructure behind persistent AI agents. At the user scale reported by Tom's Hardware, the paper calculation reaches roughly 2,000 trays, but Meta has not disclosed per-user VM specifications, server density or fleet capacity.\n\nUser tests of [Meta](https://meta.com/?ref=runtimewire)'s [Muse](https://muse.ai/?ref=runtimewire) agent found sandboxes with [two vCPUs and about 8GB of memory](https://www.tomshardware.com/pc-components/cpus/meta-muse-runs-agents-on-amd-epyc-turin-hosts-with-two-cores-and-8gb-of-memory-ai-agent-can-pass-terminal-commands-to-ubuntu-host-system?ref=runtimewire), a configuration that could fit 256 environments on a hypothetical 512-vCPU, 2TB server. Applying that density to Muse's reported user count produces an estimate of roughly 2,000 server trays. Neither the per-user allocation nor the tray count is a Meta-published service specification or fleet figure.\n\nIn its September 25th report, [Tom's Hardware](https://www.tomshardware.com/pc-components/cpus/meta-muse-runs-agents-on-amd-epyc-turin-hosts-with-two-cores-and-8gb-of-memory-ai-agent-can-pass-terminal-commands-to-ubuntu-host-system?ref=runtimewire) said Muse had reportedly passed 500,000 daily active users as of a few days earlier, citing a third-party figure. The outlet's estimate works out to about 1,953 trays at 256 users per tray, rounded to roughly 2,000. Tom's Hardware called the calculation rough napkin math: it assumes a particular server configuration and leaves actual hardware choices and overhead out of the count.\n\nBlogger Evan Hoffman and analyst Tae Kim prompted Muse to inspect its environment, according to the report. The agent's responses identified AMD EPYC 9D25 processors, Ubuntu 24.04 and Linux kernel 7.0. The sandboxes appeared to be CPU-only, with separate GPU servers handling inference, Tom's Hardware reported.\n\nA separate hands-on test measured one Muse instance at 7.7GB of RAM and 100GB of persistent storage. That report [emphasized](https://www.starkinsider.com/2026/09/meta-muse-specs-what-it-runs-on.html?ref=runtimewire) that Meta has not published or guaranteed those specifications. Meta's [September 8th launch announcement](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/?ref=runtimewire) describes Muse as running on a dedicated cloud computer, but gives no per-user CPU or memory allocation.\n\nMeta launched Muse as a personal agent that can continue working after a user closes the app and seek approval before actions such as sending email or making a purchase. The service has a free tier; [Stark Insider reported](https://www.starkinsider.com/2026/09/meta-muse-specs-what-it-runs-on.html?ref=runtimewire) that its paid Power and Maximum plans cost $20 and $100 a month, respectively. The price tiers add context to the infrastructure question, though neither the reported sandbox allocation nor the server estimate reveals Meta's actual cost to run the service.\n\nThe new compute reporting adds a different detail from RuntimeWire's [September 24th coverage](https://runtimewire.com/article/meta-s-muse-packs-roughly-70-skills-into-an-agent-with-broad-tool-access), which examined Muse's skills and access controls: how much CPU and memory a user-observed environment may consume.", "url": "https://wpnews.pro/news/meta-s-muse-may-need-2000-server-trays-at-reported-user-scale", "canonical_source": "https://runtimewire.com/article/meta-muse-linux-sandbox-ssh-security", "published_at": "2026-09-26 15:07:47+00:00", "updated_at": "2026-09-26 15:30:14.864189+00:00", "lang": "en", "topics": ["ai-agents", "ai-infrastructure", "ai-products"], "entities": ["Meta", "Muse", "Tom's Hardware", "AMD", "EPYC 9D25", "Evan Hoffman", "Tae Kim", "Stark Insider"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/meta-s-muse-may-need-2000-server-trays-at-reported-user-scale", "markdown": "https://wpnews.pro/news/meta-s-muse-may-need-2000-server-trays-at-reported-user-scale.md", "text": "https://wpnews.pro/news/meta-s-muse-may-need-2000-server-trays-at-reported-user-scale.txt", "jsonld": "https://wpnews.pro/news/meta-s-muse-may-need-2000-server-trays-at-reported-user-scale.jsonld"}}