Thanks — I’m responding to your request for recommended hardware to self-host meetings of ~100 participants at 720p. Short answer (recommended starting point) • CPU: 16–32 cores (modern Intel Xeon or AMD EPYC). Multi-core load distribution is required to avoid single-core bottlenecks. • RAM: 64 GB. • Network: 10 Gbps NIC preferred (1 Gbps minimum); ensure at least ~500 Mbps sustained symmetric bandwidth for realistic selective-subscription usage (10 Gbps for headroom). • Storage: 500 GB NVMe SSD (only required for recordings or heavy logs). • GPU: Not required unless you plan server-side transcoding or AI/video processing. Key operational recommendations • Enable simulcast and adaptive stream so each sender transmits multiple layers (720p/360p/180p) and the SFU sends only the appropriate layer to subscribers — this drastically reduces bandwidth and CPU cost. • Use selective subscription (clients only receive a few active-speaker streams) — worst-case all-to-all (everyone receiving 99 streams) is not practical. • For reliability/scale, run multiple LiveKit nodes behind a load balancer or a Kubernetes cluster with low-latency inter-node networking rather than relying on a single machine. Bandwidth guidance examples • Worst-case (everyone receives all streams): ~12.4 Gbps outbound — not realistic. • Realistic (each client receives ~4 streams at 1.25 Mbps each): ~500 Mbps total outbound. If you want, I can: • Help size the deployment for a specific participant subscription pattern (how many simultaneous high-res streams each user will receive). • Provide a sample Kubernetes resource layout (CPU/memory requests & limits) for a 3-node or single-node deployment. Which would you prefer? Sources: Going beyond a single-core | WebRTC Video Bitrate Guide | LiveKit | Bringing Zoom's end-to-end optimizations to WebRTC | Use Case: Video Conferencing | LiveKit | Announcing LiveKit Cloud