Video Encoding Cost at Scale: Per-Title Ladders, AV1, AV2

Q: Should we switch to AV2 in 2026?

For production delivery, usually not yet. AOMedia published the AV2 specification in June 2026, and its working group says software implementations come first with hardware support following. Run AV2 as an encode-side evaluation on a few titles so your pipeline is ready, and wait until your own playback logs show meaningful hardware-decode minutes before adding a full AV2 ladder.

Q: How much better is AV2 than AV1?

AOMedia working group chairs quote approximately 30% better compression than AV1 at the same quality on VMAF and PSNR, with visual gains possibly approaching 40%. Your realised saving is that gain multiplied by the share of viewing minutes on devices that decode AV2 efficiently, which starts close to zero for any new codec.

Q: What is per-title encoding?

Per-title encoding replaces a fixed bitrate ladder with one computed for each title. You encode at a grid of resolutions and bitrates, score each encode with a perceptual metric such as VMAF, and keep the cheapest bitrate that reaches each quality target. Simple content gets fewer, lower rungs; complex content gets the bits it needs.

Q: Is encoding or delivery the bigger video cost?

For any title watched more than a handful of times, delivery egress dominates. Encoding is paid once per rendition, while egress is paid on every play: a 1080p hour at 5 Mbps is about 2.25 GB on the wire. That is why slow, expensive encodes pay off for popular titles and waste money on the long tail.

Q: Does AV1 work on all devices?

No. Many recent devices decode AV1 in hardware, but older phones, TVs and set-top boxes do not, and software decode on weak mobile CPUs drains battery and drops frames. Serve AV1 only to device classes with confirmed hardware decode, keep H.264 as the universal fallback, and gate rollouts on startup time and rebuffering metrics.

Q: Why not use PSNR to compare encoders?

PSNR correlates poorly with what viewers see, and encoders can be tuned to score well on it while producing soft pictures. Use a perceptual metric such as VMAF for ladder decisions, check hard content by eye, and never compare codecs on a metric one of them was specifically tuned for.

Q: Should we use GPU or CPU video encoding?

Hardware encoders are fast and cheap per minute but usually less efficient per bit than slow software presets. Use hardware or fast presets for live and long-tail content; use slow software presets such as SVT-AV1 at high quality settings for popular titles, where saved egress across millions of plays repays the extra compute quickly.

Q: How do we estimate savings from a new codec?

Pull a week of playback logs, classify sessions by device and hardware decode support, and weight by minutes and bitrate. Multiply the codec compression gain by that hardware-decode share and by your popular-title share. Subtract added encoding and storage for the extra ladder. If the result is small, prune and per-title your ladder first.

Satyam Kumar

Back to Blog

architecture

Your Video Bill Is a Ladder Problem: Encoding Efficiency, AV1 and AV2 at Scale in 2026

By Satyam KumarNovember 9, 202612 min read

video encoding av1 av2 per-title encoding bitrate ladder vmaf svt-av1 video streaming cost cdn egress adaptive bitrate streaming hevc vs av1 video transcoding pipeline ott platform architecture dav1d video quality metrics media infrastructure

Your Video Bill Is a Ladder Problem: Encoding Efficiency, AV1 and AV2 at Scale in 2026

Frequently Asked Questions

Share this article

Twitter LinkedIn WhatsApp

Satyam Kumar

Founder & AI Architect, AppScale LLP

AI & Cloud Architect. Helping teams build systems that scale to millions.

LinkedIn GitHub

Your Video Bill Is a Ladder Problem: Encoding Efficiency, AV1 and AV2 at Scale in 2026

Frequently Asked Questions

Share this article

Comments

Leave a comment