The Alliance for Open Media's October 22 update was not a specification announcement — AV2 1.0 was published in June — but a roll call of who is already building on it, four months after the spec landed. The names on the list are the ones that matter: Google's Chrome, YouTube, Netflix, Prime Video, Meta's Facebook and Instagram, plus Intel with a new CPU encoder.

Des Keane, engineering director at YouTube, gave the timeline in the release: "YouTube AV2 streams will be available for Web as soon as early 2027." He also gave the baseline it replaces: as of March 19, 2026, more than 90% of YouTube's catalogue, weighted by watch time, is already served in AV1.

Jim Bankoski, VP of engineering for Chrome, framed browser support as "an exciting early step toward better video on the web" and said partners including YouTube, Netflix and Prime Video are exploring the format. Meta's Hassene Tmar said AV1 is now the most popular delivery format for Facebook and Instagram video and that Meta intends to carry the same investment into AV2. Intel's Iole Moccagatta pointed to SVT-AV2, the successor to the SVT-AV1 encoder that did much of the heavy lifting for AV1 production.

What the 30% actually buys

AOMedia measures AV2 at roughly 30% greater compression efficiency than AV1 — the same picture at lower bitrate, or a better picture at the same bitrate — plus work aimed at multi-stream video, real-time communication, immersive formats, higher frame rates and HDR. That is the number every streaming CFO wants.

It is also a number produced largely inside the codec's own developer circle. Independent, third-party comparisons of AV2 against AV1 and against the newest MPEG codecs are still outstanding, and the alliance's own framing is efficiency measured with objective quality metrics, not subjective viewing tests.

The bill arrives on the client

Jean-Baptiste Kempf, president of VideoLAN, put a smaller figure on the gain — improvements around 25% over AV1 are common, with individual measurements higher — and a much larger figure on the cost. Decoding AV2, he wrote, is about five times as expensive as decoding AV1. Software running on current hardware will struggle to play it in real time without optimisation aimed at the specific processor architecture.

That is why VideoLAN is already building dav2d, the successor to dav1d, the decoder that made AV1 practical on ordinary CPUs. It is also why the deployment order announced this week is telling. The first AV2 streams are web streams: a desktop browser with a fast general-purpose core and a mains cable, not a five-year-old phone on a train.

Why the transition will be slow in the places that matter

Codec shifts are now software-led and hardware-limited. AV1 shipped in 2018 and took years to become routine; it only reached 90% of YouTube's catalogue this year. AV2 has the advantage of a mature AV1 ecosystem — encoders, test streams, compliance tooling — but it starts from a decode cost that penalises exactly the devices with the least headroom: phones, tablets, smart TVs and cheap streaming sticks.

No member announced a hardware decode timeline. Until silicon arrives with AV2 blocks in it, expect a two-tier internet: AV2 for browsers and set-top boxes that can afford the cycles, AV1 for everything running on a battery. The bandwidth saving is real; it is being paid for in joules.

What to watch

Whether Netflix and Prime Video put dates on their AV2 rollouts rather than "exploring"; whether independent quality comparisons confirm the 30%; and whether dav2d lands close enough to dav1d's performance that software decoding stays viable on mid-range hardware. Until those three resolve, AV2 is a specification with excellent marketing and an unfinished client story.

Sources