TL;DR: An SFU is a server that receives one video stream from each participant and forwards copies to everyone else. It is why modern video calls scale past a handful of people. Each person uploads once regardless of group size, which keeps upload bandwidth flat while download grows with the number of participants.
Almost every video platform you have used runs on an SFU. It is the design that made group video calls practical, and it explains several behaviours users notice without understanding.
Why your laptop fan spins up in a large meeting, why upload speed matters less than you would expect, and why a weak connection degrades some participants and not others all trace back to how an SFU works.
SFU stands for Selective Forwarding Unit. It is a media server that receives one video and audio stream from each participant and forwards copies to the other participants without processing or combining them. Each person uploads once, no matter how many people are in the meeting.
The word selective matters. A good SFU does not blindly forward everything. It decides which streams each participant actually needs, and at what quality, based on what they can see and what their connection can carry.
Someone viewing a small thumbnail does not need full resolution. Someone with a poor connection should receive a lower quality version rather than a frozen high quality one. That selection logic is most of what separates a good SFU from a mediocre one.
Why does an SFU scale better than direct connections?
With direct peer connections, every participant sends a separate copy of their video to every other participant, so upload demand rises steeply with group size. An SFU means each participant uploads exactly one stream regardless of how many people are present, which is what allows meetings of hundreds rather than a handful.
| Participants | Direct: streams you send | SFU: streams you send |
|---|---|---|
| 3 | 2 | 1 |
| 8 | 7 | 1 |
| 50 | 49, in practice impossible | 1 |
Home internet connections typically have far less upload capacity than download. That asymmetry is exactly why direct connections collapse around four to six participants, and why moving the fan-out to a server solves the problem.
What is the difference between an SFU and an MCU?
An SFU forwards streams untouched, so each device receives several streams and arranges them locally. An MCU combines all streams into one composite video on the server and sends a single stream to each person. SFUs cost less to run and keep layout flexible. MCUs are far gentler on low-powered devices but fix the layout server side.
Mixing video is computationally expensive, which is why MCUs cost more to operate. Forwarding is cheap by comparison, which is why the industry settled on SFUs as the default.
The trade lands on the device. Under an SFU, your laptop decodes many streams at once, which is why fans spin up in large meetings. Under an MCU it decodes one, which is why MCUs still appear in settings dominated by older or low-powered hardware.
If you are choosing a platform rather than building one, this rarely needs to be your decision. Our guide to video call API architecture covers when it does.
Frequently Asked Questions
SFU stands for Selective Forwarding Unit. It is a media server that receives one video and audio stream from each participant and forwards copies to the others without combining them. Each person uploads once regardless of how many people are in the meeting.
An SFU forwards streams untouched, so each device receives several streams and arranges them locally. An MCU combines all streams into one composite video on the server and sends a single stream to each participant. SFUs cost less to run, while MCUs are gentler on weak devices but fix the layout server side.
Because direct peer connections require every participant to send a separate copy of their video to every other participant. Upload capacity on typical connections runs out around four to six people. An SFU moves that fan-out to a server, so each participant uploads only once no matter the group size.
No, because it forwards streams without re-encoding them. A good SFU improves the experience by selecting which streams each participant needs and at what quality, sending lower resolution to small thumbnails and to users on constrained connections rather than letting the call freeze.
Because an SFU sends your device a separate stream per visible participant, and your machine decodes each one. Decoding many streams simultaneously is demanding work. This is the trade-off SFUs make: cheaper servers and flexible layouts, in exchange for more work on the device.
Related guides
- Start with the complete video conferencing API guide
- Compare peer-to-peer, SFU, and MCU in depth
- Learn what TURN does and why it costs money
- See what WebRTC leaves out
Need large meetings without operating media servers?
Tell us your peak participant count and we will show you which request path serves it, and what the deployment options look like for regulated environments.
