The disc stopped being the unit of measurement

Four generations, four arguments · Medium

When delivery moved to streaming, the question changed from what fits on a disc to what survives a variable connection.

The Constraint That Organised Everything

For the better part of a decade, the governing question in consumer video compression was capacity. A DVD-Video disc held around 4.7 gigabytes, and MPEG-2 had to fill two hours of film inside that budget. When hacked MPEG-4 builds started circulating in the late 1990s, the target shrank further: a single 650-megabyte CD-R. Every codec decision — how aggressively to quantise, how wide to set the GOP, how much chroma subsampling to accept — answered the same underlying question. The disc was the unit. Get the bits to fit without the picture collapsing.

That constraint shaped tooling, culture and expectation in equal measure. Two-pass encoding existed specifically to manage it: the first pass analysed a source to build a complexity map, so the second pass could spend bits where the picture demanded them and starve the quiet scenes. The goal was a target file size. Not a target quality level, not a target bitrate in the abstract — a number of megabytes that happened to equal one or two optical discs.

When the Pipe Replaced the Disc

Broadband penetration in the early 2000s crossed a threshold that changed the question entirely. The constraint was no longer a fixed storage ceiling but a variable delivery floor: not how much fits, but how little you can get away with while the picture remains watchable on a screen that is almost certainly not a television. The bitrate budget was now set by a connection that could change mid-stream, and the encoder could not know in advance what that connection would be.

The industry's answer was adaptive bitrate streaming. Rather than encode a single file at a single bitrate, a title is encoded at multiple bitrates simultaneously, and the player selects among them in real time based on available bandwidth. Apple's HTTP Live Streaming specification, introduced in 2009, formalised this approach and brought it to mobile devices; MPEG's Dynamic Adaptive Streaming over HTTP, known as MPEG-DASH, followed in 2012. Both methods rely on the same principle: segment the stream into short chunks, and serve whichever quality tier the connection can sustain at that moment.

The codec consequences were significant. A keyframe must begin each segment cleanly, because the player switches tiers at segment boundaries and cannot rely on reference frames from a previous stream. This forces keyframe placement to align with segment timing rather than with the content's own pacing — a constraint that had nothing to do with how the picture looks and everything to do with how it travels. Two-pass encoding, optimised to hit a file-size target, became less central; constant-quality modes and per-title optimisation, which ask a different question entirely, grew in importance.

What the Codec Negotiates Now

H.264/AVC, standardised in 2003, arrived almost exactly at the inflection point, which goes some way toward explaining its dominance. It was efficient enough to serve watchable video at the bitrates a residential broadband connection could deliver, and hardware decode support arrived quickly enough that mobile devices could play it without draining a battery in the process. The Moving Picture Experts Group had designed it for broadcast and disc, but it turned out to suit the pipe as well.

What streaming made visible was that lossy compression is not just about what you throw away — it is about when you can throw it away and still have the player recover in time. Generation loss, the compounding degradation of re-encoding an already-lossy source, became a practical concern at scale: a title transcoded multiple times on the way to a viewer accumulates errors in ways that a single disc master did not. The question of how much to discard had acquired a new dimension. The disc had told the encoder exactly how much room it had. The stream tells it only what the connection can bear right now, and that changes by the second.

The Alliance for Open Media's AV1, published in 2018, was designed with this environment as its primary context: royalty-free, efficient at low bitrates, and built to run inside the browser and the streaming appliance rather than off a laser-read platter. The disc had organised a decade of decisions. What replaced it organised the next one.