WebRTC Mesh vs SFU vs MCU: Video Call Topologies Compared

Mesh, SFU and MCU are the three ways to connect people in a WebRTC video call. Mesh links every device to every other, an SFU server forwards each stream without decoding it, and an MCU server mixes all streams into one. Your room size mostly decides which you need.

TL;DR:

• Mesh: every participant connects to every other, so N people create N x (N-1) streams. It holds up for 3 or 4 people and breaks down past five.
• SFU (Selective Forwarding Unit): each participant uploads once, and the server forwards that stream to the others without re-encoding it. Fora Soft prices a self-hosted SFU at about $0.0008 to $0.003 per participant-minute.
• MCU (Multipoint Control Unit): the server decodes every stream, mixes one picture and re-encodes it, at roughly 4 to 10 times an SFU's compute and about $0.008 to $0.025 per participant-minute.
• Room size maps almost directly to topology: P2P for 1:1, SFU from small groups to hundreds, and SFU plus LL-HLS or a CDN for audiences in the thousands.
• AIOZ Stream doesn't run call topologies. It covers the broadcast side: RTMP ingest, HLS playback and a Low Latency Mode (Beta) for interactive streams.

Updated Oct 19, 2026: replaced the line about low-latency live streaming "once it ships" with AIOZ Stream's documented live path and Low Latency Mode (Beta), added a topology diagram and a room-size table, and shortened the opening and sentences.

What is the difference between mesh, SFU and MCU?

All three answer the same question: how do N people in a live call see and hear each other? Mesh sends media straight between your devices, an SFU puts a server in the middle that only routes packets, and an MCU puts a server in the middle that builds one combined video for everyone.

These topologies often get filed under "P2P video," which invites a mix-up, because they're about real-time, many-to-many calls. The content-delivery mesh in our CDN edge node guide is a different thing: cached copies of one broadcast fanning out to many viewers.

Mesh: the math that breaks it

Full mesh is the simplest topology to reason about and the fastest to outgrow. Every participant opens a direct connection to every other participant, and no server touches the media. With N people, that's N x (N-1) streams, because each person both sends to and receives from everyone else.

A 4-person call means 4 x 3 = 12 streams, and your device runs 3 uploads and 3 downloads at once. At 1 Mbps per stream, that's 3 Mbps of upload plus the CPU to encode three outgoing copies. Fora Soft puts the ceiling plainly: mesh "holds up for 3-4 people on decent devices; past five it burns mobile batteries and saturates uplinks." Growth is quadratic, so doubling the room roughly quadruples the traffic. With no server, mesh costs you nothing to host, which makes it the right default for 1:1 and very small calls.

SFU: forward, don't decode

An SFU breaks the quadratic growth by putting a server in the middle that doesn't process video. You upload your stream once, and the SFU forwards that same encoded stream to every participant who needs it. Because the server routes packets instead of decoding them, its CPU cost stays low and latency stays close to a direct connection.

Most SFUs add simulcast: each sender uploads a few quality layers, and the server picks the layer that suits each receiver. That way, one meeting can serve a phone and a 4K monitor without server-side transcoding. Fora Soft's cost table puts a self-hosted SFU at about $0.0008 to $0.003 per participant-minute, which is why SFU is the default from small groups up to several hundred people.

MCU: mix everything into one stream

An MCU takes the opposite approach and puts the heavy lifting on your server. It decodes every incoming stream, composites them into one picture (usually the familiar grid), re-encodes it and sends one stream to each viewer. That's continuous, expensive video processing for the length of your call. Fora Soft estimates roughly 4 to 10 times an SFU's compute per participant, or about $0.008 to $0.025 per participant-minute.

MCU survives for a narrower case, when the receiving device is your bottleneck. A low-bandwidth dial-in, a weak mobile link or a thin client that can't decode several streams gets exactly one ready-made picture. You pick it for those endpoints, not for quality or cost.

Which topology fits your room size?

Peak participants per room decides more than any other factor, and Fora Soft's sizing table maps it like this:

Past about a thousand viewers, your setup becomes a hybrid. An SFU handles the interactive stage, and LL-HLS or plain HLS over a CDN carries the stream to everyone else. WebRTC doesn't cache, so large audiences need HTTP delivery somewhere in the pipeline, as our HLS vs WebRTC guide explains. Our overview of how the main streaming protocols compare covers that hybrid alongside RTMP and SRT ingest.

Do Zoom and Google Meet use SFU or MCU?

SFU-style forwarding. Fora Soft notes that nearly every modern conferencing product, including Zoom, Teams and Meet, forwards streams selectively on the live path rather than mixing one canvas on the server. Zoom calls its media servers "multimedia routers," which describes the job. MCU-style mixing shows up mainly where an endpoint, such as a phone dial-in, can only handle one stream.

How AIOZ Stream handles it

AIOZ Stream doesn't host video calls, so none of these three topologies describes our platform. The docs don't cover an SFU, MCU or WebRTC path. What we do cover is the audience side of the hybrid pattern above: you send a program feed over RTMP from OBS Studio or PRISM Live Studio, and viewers watch over HLS.

For interactive broadcasts such as Q&A, auctions or gaming, the dashboard player view has a Low Latency Mode (Beta) that reduces the delay between you and your audience. The docs don't publish a latency figure for it. You can also forward the same live stream to other platforms. The multicast reference documents this request (keys and URL are placeholders):

curl --location --request POST 'https://api.aiozstream.network/api/live_streams/multicast/stream_key' \
  --header 'stream-public-key: your_public_key_here' \
  --header 'stream-secret-key: your_secret_key_here' \
  --header 'Content-Type: application/json' \
  --data-raw '{ "multicast_urls": [ "rtmp://a.rtmp.youtube.com/live2/abcd-abcd-abcd-abcd" ] }'

The DePIN network behind AIOZ Stream is also a different layer from mesh, SFU and MCU. It distributes cached copies of already-encoded video across operator-run nodes, while call topologies route live media between participants before anything is cached. Next step: set up a broadcast with the Create Livestream guide.

Frequently Asked Questions

What's the difference between mesh, SFU and MCU in one line each?
Mesh connects every participant directly with no server, an SFU forwards streams without decoding them, and an MCU decodes, mixes and re-encodes all streams into one per viewer.

Why does mesh fall apart past 4 or 5 participants?
Total streams grow as N x (N-1), so upload bandwidth and CPU on each device climb quickly as the room grows, saturating uplinks and draining phone batteries.

Is MCU ever the better choice?
Yes, when your receiving device can't handle several incoming streams, such as a low-bandwidth dial-in or a thin client; otherwise quality and cost both favor an SFU.

Does AIOZ Stream use mesh, SFU or MCU?
No. AIOZ Stream's documented live path is RTMP in and HLS out, with a Low Latency Mode (Beta) for interactive broadcasts, and on-demand video plays over HLS or DASH.

Is this the same "P2P" as AIOZ's DePIN network?
No. The DePIN network distributes cached copies of encoded video across operator-run nodes, while mesh, SFU and MCU route live media between call participants.

Related Reading

• HLS vs WebRTC: the protocol tradeoff these topologies sit inside
• What Is Low-Latency HLS (LL-HLS): the delivery layer large broadcasts pair with an SFU stage
• Glass-to-Glass Latency: What It Is and How to Measure It
• Live Streaming on AIOZ Stream: The Complete Guide: RTMP ingest, restreaming and archiving
• What Is a CDN Edge Node: the content-delivery meaning of "P2P"
• AIOZ Stream Explained: how upload, storage and delivery fit together
• libp2p: The Networking Layer Underneath IPFS: peer connections and NAT traversal in another P2P stack

References

• Fora Soft: SFU vs MCU vs P2P, Choose a WebRTC Architecture: stream math, cost per participant-minute and the room-size table
• Fora Soft: P2P vs MCU vs SFU for Video Conference Apps: selective forwarding in Zoom, Teams and Meet
• Fora Soft: Simulcast: how SFUs serve different quality layers
• AIOZ Stream docs, Viewing and Sharing and Livestream Multicast: Low Latency Mode (Beta) and restreaming