Why P2P Scales
Scale to Millions of Users
Without Millions in Bills
Peer-to-peer architecture means users carry the bandwidth load, not your servers. Here's why that changes everything.
The Economics of P2P
In traditional video apps, you pay for every bit of data. In P2P, users share that load.
Traditional SFU/MCU
High infrastructure costs
Per month for 1,000 concurrent users
All video routes through your servers
Costs grow with every user
P2P Mesh Network
Minimal infrastructure
Media flows directly between users
Users provide their own bandwidth
Only signaling costs grow
What We Actually Pay For
Our infrastructure has three components. Only one has meaningful cost.
Signaling Server
Coordinates peer connections via WebSocket. Handles room state, user presence, and SDP/ICE exchange.
- Lightweight WebSocket messages
- Text only (no media)
- Single server handles 100K+ users
- Memory: ~KB per connection
P2P Mesh
Video, audio, and chat flow directly between participants. Your server never sees this traffic.
- Users provide bandwidth
- DTLS-SRTP encryption
- Data channels for chat
- No server involvement
TURN Relay
Fallback for ~10-20% of connections when NAT traversal fails. The only real scaling concern.
- Only used when P2P fails
- Self-host or use providers
- $0.02/GB typical cost
- Mitigatable with optimization
Cost at Scale
See how costs compare as you grow. Move the slider to explore different scales.
Media servers + bandwidth + scaling infrastructure
Signaling server + TURN fallback (15% usage)
When P2P Fails: TURN Relay
About 10-20% of connections can't establish direct P2P due to restrictive NATs. These require TURN relay servers.
TURN Cost Breakdown
| Scenario | Connections | TURN Usage (15%) | Bandwidth/mo | Cost @ $0.02/GB |
|---|---|---|---|---|
| 100 users in calls | 100 | ~15 | ~50 GB | ~$1 |
| 1,000 users in calls | 1,000 | ~150 | ~500 GB | ~$10 |
| 10,000 users in calls | 10,000 | ~1,500 | ~5 TB | ~$100 |
| 100,000 users in calls | 100,000 | ~15,000 | ~50 TB | ~$1,000 |
Mitigation Strategies
- Force STUN-only first, TURN as fallback
- Use TCP TURN (cheaper, higher latency)
- Reduce video quality over TURN
- Self-host TURN to control costs
Why 8 Participants Max?
P2P mesh creates connections between every participant. This limits group size but eliminates server costs.
Mesh Topology
Formula: n(n-1)/2 peer connections
8 participants = 28 connections. Beyond this, mesh becomes impractical and an SFU would be needed.
Perfect For
- 👥1-on-1 conversations
- 👨👩👧👦Small team standups (2-5)
- 🎮Gaming groups (4-8)
- 📚Study groups & tutoring
- 🏢Large webinars (50+)
- 🎪All-hands meetings
Summary: P2P vs Traditional
| Feature | Traditional SFU | P2P Mesh (Nexus) |
|---|---|---|
| Media Server Cost | High ($$$) | None ($0) |
| Bandwidth Cost | All on you | Users pay own |
| Scaling Cost | Linear growth | Near-zero |
| Max Participants | Unlimited | 8 per room |
| Latency | Higher (server hop) | Lower (direct) |
| Privacy | Server sees all | True E2E |
| Reliability | Depends on server | Depends on peers |
Build the next big thing
without the big bills
Whether you're bootstrapping or raising funds, P2P gives you runway that traditional architectures can't match.