Transcoding is converting a source video into the multiple renditions and codecs the platform needs to deliver. The source — a high-bitrate mezzanine for VOD or a contribution feed for live — is decoded and then re-encoded into every rung of the encoding ladder, in every codec the device matrix requires (H.264, HEVC, AV1), with the corresponding audio and any required processing.
Transcoding is the most compute-intensive stage of the pipeline. For VOD it can run offline, in parallel, with slow multi-pass encodes that maximize quality per bit; for live it must run in real time with a fixed latency budget, which constrains how much optimization is possible. This split drives the build of a transcoding farm — cloud, on-prem, or hybrid — sized to throughput and balanced against the cost of compute.
Because transcoding multiplies one source into many outputs (renditions × codecs × audio × subtitles), it is a major cost and operational center, and smart encoding (per-title, context-aware) is largely about doing less of it without losing quality.
