A live stream reaches viewers noticeably after the event it shows, and the delay is not a fault. It is built deliberately, in several stages, each buying stability at the cost of time.
Encoding costs time before anything is sent
Compressing video requires examining frames in relation to one another, which means the encoder holds a short window of footage before it can produce output for the beginning of that window.
Higher compression efficiency requires a longer window, so the settings that reduce bandwidth also increase delay, and the two cannot be optimised independently.
Hardware encoders shorten this considerably, which is why dedicated equipment produces lower latency than software running alongside everything else on a computer.
Delivery is built from segments
Most streaming delivers video as a series of short files rather than a continuous flow, because that structure works with ordinary web infrastructure and scales to large audiences cheaply.
A segment cannot be sent until it is complete, so a segment length of several seconds guarantees at least that much delay before delivery even begins.
Shortening segments reduces latency and increases overhead, since each one carries its own request and header, which is why the choice is a compromise rather than a setting to minimise.
The player holds a buffer on purpose
Players accumulate several segments before displaying anything, so that a momentary network problem is absorbed silently instead of stopping playback.
That reserve is the difference between a stream that continues through a brief interruption and one that stalls visibly, and it is the largest single component of total delay.
Low-latency modes reduce the reserve deliberately, which is why they are more prone to buffering on imperfect connections and are not enabled by default.
Delay is sometimes added deliberately
Broadcasts that must be moderated introduce additional delay so that a person can intervene before material reaches an audience, which no automatic system can do after transmission.
Interactive formats add their own consideration, since chat arrives with lower delay than video, and a presenter reading responses is reacting to comments about something already past.
Why viewers do not all see it together
Each viewer's player manages its own buffer according to its own connection, so two people watching the same stream can be several seconds apart from one another.
That spread is why live formats depending on simultaneity, such as competitive play or timed participation, require infrastructure quite different from ordinary streaming delivery.