Viewers of a live stream regularly see chat react to something before it appears on screen. The two are travelling over the same connection by very different routes, and one of them is far slower.
Chat has almost nothing to do
A chat message is a few dozen bytes. It is sent, relayed and displayed with no processing that takes meaningful time, arriving in a fraction of a second.
There is no compression stage, no buffering requirement and no need to accumulate anything before display, because a message is complete the moment it is written.
The result is that text moves at close to the speed of the network itself, which is the fastest anything in the system can travel.
Video is delayed at every stage
The stream is encoded, which requires holding frames; packaged into segments, which requires waiting for a segment to complete; then buffered by the player before display.
Each stage adds seconds, and they accumulate. A stream can easily run somewhere between ten and thirty seconds behind the moment it was captured.
None of this delay applies to a chat message, so the gap between the two is the total of everything the video had to pass through.
The delay is different for every viewer
Viewers receive the stream from different servers, on different connections, with players making independent buffering decisions based on local conditions.
Two people watching together can therefore be several seconds apart, while both see the same chat at the same instant, since chat is not buffered per viewer.
This is why chat becomes incoherent during fast events. Participants are reacting to different moments while reading a single shared timeline.
Spoilers are a structural consequence
Anything time-sensitive, particularly sport or competitive events, is revealed in chat before it reaches viewers running further behind.
Platforms with alerts and notifications make this worse, since those systems are also fast and are not synchronised to any individual's playback position.
Some services address it by delaying chat to approximate the stream, which improves coherence at the cost of making conversation feel sluggish.
Why hosts feel it most
A presenter reading chat is responding to messages written about video from many seconds ago, which makes the exchange feel disjointed from both sides.
Experienced hosts compensate by acknowledging the lag explicitly and by not treating chat as a real-time reaction to the current moment.
Reducing stream latency shortens the gap and costs stability, which puts interactive broadcasters back on the same trade every live stream has to make.