Estimating 100M DAU infra for Meta Muse: 1-4GW of power, tiny $3B sandbox layer
SuB8u · x · 2026-09-28
Freda Duan published a rough estimate of the infrastructure required to serve 100M DAU on Meta's Muse:
- Baseline needs 1 GW of power, of which only 0.1 GW comes from the CPU/VM layer; depending on reasoning-equivalent model calls per user per day, 3-4 GW is plausible — possibly explaining Meta's rumored 7-10 GW compute addition next year.
- The sandbox layer is far smaller than expected: under $1B of CPU content and $2B of DRAM content.
- Muse runs on two distinct infrastructure layers: a per-user VM/sandbox (2 vCPUs, 8 GB RAM, 100 GB persistent storage) and model inference routed through Meta's external inference infrastructure.
The author flags many moving assumptions and invites pushback.
More from Infra
- Gemini 3.5 Flash now runs on air-gapped Google Distributed Cloud, offline up to 12 months — shashib · 2026-11-16
- Gemma.c: full Gemma 1/2/3 multimodal inference in a single C file, ~72 tok/s on CPU at INT8 — This-Purchase-3325 · 2026-11-16
- Sarvam AI Launches Token Factory: India-Hosted Inference for Open Models — itsOmSarraf_ · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Jailbroken PS4 runs a 9B open-source model at up to 43 tok/s by talking directly to its 2013-era GPU — AIFlow_ML · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16