Originally published at v100.ai
At V100, we build AI video infrastructure entirely in Rust. 20 microservices. 0.01ms server processing. 220,000+ requests per second. Post-quantum encryption on every call.
This post is a deep dive into our architecture and the engineering decisions behind it.
Why Rust for Video Infrastructure?
Video infrastructure has unique constraints that make Rust the ideal choice:
- Zero GC pauses — Real-time video processing can't tolerate garbage collection stops
- Memory safety — Buffer overflows in media pipelines are a security nightmare
- Performance — Our gateway processes requests in 10 microseconds, not 10 milliseconds
- Small binaries — Our meeting signaling server is a 2MB binary serving WebRTC at scale
Our Stack
| Component | Technology |
|---|---|
| Web Framework | Axum + Tokio |
| Database | PostgreSQL + TimescaleDB |
| Cache | Redis + Cachee (in-process) |
| Media | FFmpeg (sidecar) |
| Crypto | ML-KEM-768 + ML-DSA-65 (post-quantum) |
| Infra | Docker + AWS ECS Fargate |
The Numbers
- 20 Rust microservices — gateway, AI, transcription, video, conferencing, billing, broadcasting
- 0.01ms server processing latency
- 220K+ RPS sustained throughput
- 263ns pipeline latency
- 938 tests with zero failures
- 40+ languages for real-time transcription
- 7 platforms for social publishing (YouTube, TikTok, Instagram, LinkedIn, X, Facebook, Vimeo)
Key Services
- Gateway (Axum) — JWT auth, rate limiting, CSRF/SSRF protection
- AI Orchestration — Claude/Gemini proxy, streaming, compliance
- Transcription — Deepgram + Whisper, word-level timestamps, 40+ languages
- Meeting Signaling — WebRTC SDP/ICE, DashMap concurrency, 2MB binary
- v100-turn — Full broadcast platform: ABR, DRM, DVR, spatial audio, deepfake detection
Try It
V100 has a free tier — v100.ai/pricing
API docs at docs.v100.ai
Read the full technical deep-dive at v100.ai/blog/sfu-architecture-200-participants
V100 is built by H33.ai — post-quantum security for production systems.
Top comments (0)