Saturday, October 24th 2026

YouTube Sets Early 2027 Target for AV2 Streams on the Web

The Alliance for Open Media (AOMedia) shared an update on AV2 this week, and the clearest date in it comes from YouTube. Des Keane, Engineering Director at YouTube, confirmed the platform will begin streaming in AV2, which is the royalty-free successor to AV1 that AOMedia says offers roughly 30% better compression. In practice, that means either sharper video at the same bitrate or the same quality using less bandwidth, and the codec also adds support for multi-stream and multi-view video, higher frame rates, and improved HDR. Keane put it plainly: "YouTube AV2 streams will be available for Web as soon as early 2027."

Chrome is the other half. Jim Bankoski, VP of Engineering for Google Chrome, said AV2 playback has been demonstrated in the browser, though Google hasn't said which Chrome version will get it or whether it will be on by default. That demo dates back to CES 2026, where Google streamed AV2 videos from YouTube to a gaming laptop at 1080p and 24 FPS using a custom Chrome build with the reference decoder. Netflix and Prime Video are also exploring the format, but neither has shared a timeline. Meta wants to repeat its AV1 rollout with AV2, and Intel is working on SVT-AV2.
In terms of hardware, it looks like no consumer GPU or TV supports AV2 hardware decoding yet, so early playback will probably lean on the CPU. According to Videocardz, VideoLAN's dav2d decoder should help there, although VideoLAN says AV2 decoding is roughly five times more complex than AV1. Over 90% of YouTube's catalog, weighted by watch time, was already available in AV1 as of March 2026, so it's reasonable to expect AV2 to take a similar path, starting in software on the web and moving to dedicated silicon later.
Sources: Alliance for Open Media, Videocardz
Add your own comment

32 Comments on YouTube Sets Early 2027 Target for AV2 Streams on the Web

#1
Dr. Dro
It's interesting as a technology preview of sorts, I guess. With no decoding capability offered by any IHV even in the latest generation hardware, the usefulness is relatively limited. AV1 will probably not be phased out as the flagship codec any time soon - VVC is bound by high complexity and patents, AV2 by sheer newness, both suffer from no fixed function enc dec as of today.
Posted on Reply
#2
jdrch
Dr. DroIt's interesting as a technology preview of sorts, I guess. With no decoding capability offered by any IHV even in the latest generation hardware, the usefulness is relatively limited. AV1 will probably not be phased out as the flagship codec any time soon - VVC is bound by high complexity and patents, AV2 by sheer newness, both suffer from no fixed function enc dec as of today.
My thoughts exactly. This is useful only with hardware decoders. Too bad Google are terrible at hardware partnerships. Unless they add AV2 decoding to certified Android requirements as they did with their Widevine DRM.
Posted on Reply
#3
huggi
Higher quality at the same bitrate would be nice since Youtube's current video quality can be shit depending on the content. But bandwidth is expensive so I have no doubt that lower bitrate will be one of, if not the, main focus.
Posted on Reply
#4
Dr. Dro
huggiHigher quality at the same bitrate would be nice since Youtube's current video quality can be shit depending on the content. But bandwidth is expensive so I have no doubt that lower bitrate will be one of, if not the, main focus.
Yeah, YouTube deliberately leaves bitrate low for free users. Some videos have offered a high(er) bitrate mode in some videos for some time now, as long as you have a Premium subscription. But "1080p Premium" mode is only available in videos being served in VP9, if a 1440p or 4K version is available, then it is never served.



AV2 should really help conserve bandwidth, my own testing with VVC has yielded visually lossless DVD quality video (480p 24) at essentially mp3 bitrates (400k ish). This was early last year though, encoders probably improved since, and probably will not take several hours to encode a minute with a 14900K-equivalent CPU :fear:
Posted on Reply
#5
Quitessa
It's getting to the point that the Audio bitrate will be higher than the Video bitrate.
Posted on Reply
#6
Nhonho
I think that, using AI in data centers, it's possible to create video and audio codecs that are far more efficient than the best ones available today.

Before AI, programmers had to be highly skilled in mathematics and statistics—and also very creative—to improve something that was already near perfection. Today’s AI can go much further, by way of trial and error.
Posted on Reply
#7
Dr. Dro
NhonhoI think that, using AI in data centers, it's possible to create video and audio codecs that are far more efficient than the best ones available today.

Before AI, programmers had to be highly skilled in mathematics and statistics—and also very creative—to improve something that was already near perfection. Today’s AI can go much further, by way of trial and error.
It's not that simple, and there's a whole patent minefield that has to be carefully navigated. It's not really feasible to vibe code a hyper-efficient video codec, at least not yet.
Posted on Reply
#8
Logic
DrUsmanSaleemThe Alliance for Open Media (AOMedia) shared an update on AV2 this week, and the clearest date in it comes from YouTube. Des Keane, Engineering Director at YouTube, confirmed the platform will begin streaming in AV2, which is the royalty-free successor to AV1 that AOMedia says offers roughly 30% better compression. In practice, that means either sharper video at the same bitrate or the same quality using less bandwidth, and the codec also adds support for multi-stream and multi-view video, higher frame rates, and improved HDR. Keane put it plainly: "YouTube AV2 streams will be available for Web as soon as early 2027."

Chrome is the other half. Jim Bankoski, VP of Engineering for Google Chrome...
Meanwhile recent 'updates' to Chrome/Chromium and FireFox have dropped support for older Intel IGP:
Yes the CPU did decoding, then handed rendering, compositing etc off to the GPU.
Now:
The CPU is forced to do everything... And simply does not have the oomph to 'do it all', obsoleting older hardware independent of OS.
ie: Linux or Windows; your hardware is now 'too old' to surf.

Specific eg:
Client's old G45 system running Q4OS Trinity and Chromium, which used to surf just fine is struggling and Youtube is 'not happening' anymore.

Proof that the iGPU is still capable of rendering etc:
GTK-Pipeviewer plays Youtube video just fine using CPU decode, but IGP rendering etc as it always used to.

You might ask Jim Bankoski and co about that plz!?
Posted on Reply
#9
AsRock
TPU addict
Dr. DroYeah, YouTube deliberately leaves bitrate low for free users. Some videos have offered a high(er) bitrate mode in some videos for some time now, as long as you have a Premium subscription. But "1080p Premium" mode is only available in videos being served in VP9, if a 1440p or 4K version is available, then it is never served.



AV2 should really help conserve bandwidth, my own testing with VVC has yielded visually lossless DVD quality video (480p 24) at essentially mp3 bitrates (400k ish). This was early last year though, encoders probably improved since, and probably will not take several hours to encode a minute with a 14900K-equivalent CPU :fear:
I use 4k on youtube all the time when available, if browsing on the PC it seems to default to 1080 but on my roku it stays as set unless the video was not done in 4k..

And do not have youtube account.
Posted on Reply
#10
810Cat
Their AV1 doesnt work properly and they're already putting av2? Bruh.
Posted on Reply
#11
dgianstefani
TPU Proofreader
huggiHigher quality at the same bitrate would be nice since Youtube's current video quality can be shit depending on the content. But bandwidth is expensive so I have no doubt that lower bitrate will be one of, if not the, main focus.
It's intentional so YT can sell "enhanced" 1080p with Premium.
Posted on Reply
#12
GeoffreyA
Dr. DroAV2 should really help conserve bandwidth, my own testing with VVC has yielded visually lossless DVD quality video (480p 24) at essentially mp3 bitrates (400k ish). This was early last year though, encoders probably improved since, and probably will not take several hours to encode a minute with a 14900K-equivalent CPU :fear:
The problem with VVenC, the only practical VVC encoder, is that a lot of detail is lost, and Fraunhofer's work on the encoder seems to have lessened. SVT-AV1 is doing much better in detail these days, and speed is excellent.
Posted on Reply
#13
silentbogo
NhonhoI think that, using AI in data centers, it's possible to create video and audio codecs that are far more efficient than the best ones available today.
You don't have to. It's all there. FrameGen and DL upscaling were first demoed quite a few years ago as a way to add FPS not to games, but to videos. Same with several denoising algos, which later morphed in what we know today as DLSS.
Posted on Reply
#14
GeoffreyA
NhonhoI think that, using AI in data centers, it's possible to create video and audio codecs that are far more efficient than the best ones available today.

Before AI, programmers had to be highly skilled in mathematics and statistics—and also very creative—to improve something that was already near perfection. Today’s AI can go much further, by way of trial and error.
Patent nightmares aside, it will go in this direction. Traditional block-based encoding has hit critical mass since x264, and the gains since then have been tightening up the creases and diminishing returns.
Dr. DroAV1 will probably not be phased out as the flagship codec any time soon - VVC is bound by high complexity and patents
VVC missed the train and will now wait at the station indefinitely, it seems.
Posted on Reply
#15
NorthByte
The 2027 timeline for initial web deployment makes total sense from Google's CDN bandwidth economics, but client-side adoption is going to face the classic silicon lag. We are only now in an era where mainstream budget streaming boxes and TV SoCs (like Amlogic S905X4/Y4 and MediaTek Pentonic series) reliably include dedicated fixed-function AV1 hardware decoding as standard.
AV2 targets roughly 25-30% compression efficiency gains over AV1, but the compute complexity required for software decoding will make CPU fallback completely impractical on low-power Android TVs, smart sticks, and mobile hardware.
Google will almost certainly replicate the staged rollout they used with VP9 and AV1: start with 4K/8K desktop browser streams where discrete GPUs and modern desktop CPUs can brute-force or assist the decoding, while continuing to serve AV1 and VP9 to living room and mobile clients until fabless semiconductor vendors bake fixed AV2 ASIC blocks into their 2027-2028 reference designs.
Posted on Reply
#16
john_
- I just bought a brand new AV1 capable graphics card to play all those beautiful Youtube videos.
- I have bad news for you.
Posted on Reply
#17
NorthByte
Haha, thankfully that card is safe for a very long time! Even when Google begins pushing AV2 for initial 4K/8K desktop Chrome sessions around 2027, the immense catalog of existing VP9/AV1 transcodes on YouTube's edge caches isn't going anywhere.
Hardware rollouts always have massive silicon inertia. For living room devices, streaming sticks, and lean media players, AV1 is going to remain the practical baseline standard through the end of the decade.
Posted on Reply
#18
chrcoluk
huggiHigher quality at the same bitrate would be nice since Youtube's current video quality can be shit depending on the content. But bandwidth is expensive so I have no doubt that lower bitrate will be one of, if not the, main focus.
Youtube aka google I have noticed have now jumped on the bandwagon of forgetting previously set target resolution and defaulting to 480p or 720p (if lucky), scraping for pennies as one of the richest countries in the world. Combine this is with the aggressive anti ad blocker stance, and the super weird embedded videos UI the web keeps regressing. (yeah lets move all the control elements around, you click settings cog top right only for menu to appear at bottom lolz).
Posted on Reply
#19
kondamin
nice, shame they don't sync with hardware manufacturers before starting with a mass rollout.
Now their saved bandwidth will result in users extra power use.
Posted on Reply
#20
ZeDestructor
kondaminnice, shame they don't sync with hardware manufacturers before starting with a mass rollout.
Now their saved bandwidth will result in users extra power use.
It's not actually that problematic in the long run. Sure, it's not a good idea for mobile (which is why older codecs live for a while, like VP9 is right now alongside AV1, and the backup H264 720p stream), but for desktops and (plugged in) laptops software decoders can be rather fast. For example, the dav1d 1.4.2 AV1 decoder (current is 1.5.4 and that has even more performance improvements) on a 10W i7-8565 with 4 Skylake cores decodes 4K AV1 at a very healthy 71-77fps. That's an 8 year old CPU (I tried to find older, but they were only tested on older, slower versions of dav1d), on a 10 year old architecture and powercapped to a truly miserably low power level and it's still enough to handle 4K60 decoding entirely in software. With a few cycles to spare for a GUI, networking and input handling!

The dav2d decoder will obviously not be as fast - both because of how early things are and how AV2 is more complex, but by the time AV2 becomes available on most of the youtube library, the 10 year old machines then should be fast enough to handle it entirely in software.

What I actually don't like is how we're going to see et another reduction in bandwidth instead of an increase in image quality, in a world where bandwidth gets cheaper every year on average. And don't even get me started on the audio side of things - audio should be lossless by default in 2026, yet here we are.
Posted on Reply
#21
AddSub
Something that is always missed by these "new thing reduces bandwith requirments by 10%!" announcements is that the computational savings are offloaded onto YOU. Always. Yes, of course, if you wait a few years, or much longer considering the pace of hardware development and progress has now almost completely stalled, you will EVENTUALLY get a GPU/CPU that has native hardware decode extensions to handle the new format with no or little computational penalty. Until then, your GPU or most likely CPU is going to be chocking on that, to quote the marketing here: "30% better compression".

Yes, you save few mb here or gb there, on longer streams bandwith-wise, but that "30% better compression" has to be eventually decompressed client side, and that client is most likely running a Ryzen from few years back that has no clue how to handle the new shiny codec.

h264 for all its bandwidth "hungryness" can be natively decoded on freaking GeForce 8600 GT from 2007! Save a few mb/gb on some YouTube stream on Sumerian Nibiru alien goobla-goobla while your CPU/GPU chokes decoding those AV2 "bandwidth savings" or have h264 do it at 0.1% hardware load while you gooble up few extra mb/gb of bandwith... I would always choose latter.

It's all very 21st century "poop flows downstream river, their problem down there" type of engineering. Yeah, we Google and our clients save on bandwidth costs, network side, while millions if not billions of users have their CPUs/GPUs pegged until Intel/Nvidia/whomever decides to add native extensions. Sigh... its all so very....

...
..
.
Posted on Reply
#22
dgianstefani
TPU Proofreader
AddSubSomething that is always missed by these "new thing reduces bandwith requirments by 10%!" announcements is that the computational savings are offloaded onto YOU. Always. Yes, of course, if you wait a few years, or much longer considering the pace of hardware development and progress has now almost completely stalled, you will EVENTUALLY get a GPU/CPU that has native hardware decode extensions to handle the new format with no or little computational penalty. Until then, your GPU or most likely CPU is going to be chocking on that, to quote the marketing here: "30% better compression".

Yes, you save few mb here or gb there, on longer streams bandwith-wise, but that "30% better compression" has to be eventually decompressed client side, and that client is most likely running a Ryzen from few years back that has no clue how to handle the new shiny codec.

h264 for all its bandwidth "hungryness" can be natively decoded on freaking GeForce 8600 GT from 2007! Save a few mb/gb on some YouTube stream on Sumerian Nibiru alien goobla-goobla while your CPU/GPU chokes decoding those AV2 "bandwidth savings" or have h264 do it at 0.1% hardware load while you gooble up few extra mb/gb of bandwith... I would always choose latter.

It's all very 21st century "poop flows downstream river, their problem down there" type of engineering. Yeah, we Google and our clients save on bandwidth costs, network side, while millions if not billions of users have their CPUs/GPUs pegged until Intel/Nvidia/whomever decides to add native extensions. Sigh... its all so very....

...
..
.
So what, technological progress should stop after h264? Ofc the software comes first and hardware later.
Posted on Reply
#23
AddSub
dgianstefaniSo what, technological progress should stop after h264? Ofc the software comes first and hardware later.
Difference is that in the decades past sometimes the fresh new encoding/decoding tech was followed by hardware extensions relatively quickly, sometimes within the 12mo-18mo period in the breakneck 1990-2010 era. So it was an easy-peasy era of: "Hey, this new shiny gooblaVersion10 encoder is kinda pegging my CPU right now but it dosent matter, next year Intel/AMD will include native extensions AND that new CPU will be x1.5-x2.0 faster IPC wise than my current one anyway!"

Nowadays? Well you see how it is! Not only are generational hardware releases spaced much more apart, there is no guarantee AV2 will be baked in...

So yeah, release pace of new encoders and standards were fine when hardware was being released at breakneck pace. Now, AV2 hardware decoding is planned on consumer GPUs/CPUs.... when? early 2030s? LOL.

It might be eternal 2005 in well insulated from reality Silicon Valley, and Avatar: The Last Airbender just aired, but most of the planet is firmly in 2026 and most of that planet is running decade old Intel hardware and GPUs from the Maxwell era...

No, this benefits Google first where they will recoup massive bandwith saving costs for themselves and their clients. You, the average Joe with a GTX 1080 and a Ryzen 3600x? LOL. Whats a few watts extra of power consumption anyway....
Posted on Reply
#24
cvaldes
AddSubSo yeah, release pace of new encoders and standards were fine when hardware was being released at breakneck pace. Now, AV2 hardware decoding is planned on consumer GPUs/CPUs.... when? early 2030s? LOL.
My guess is that AV2 hardware decoding support will come to smartphone SoCs before mainstream consumer PC GPUs.

Remember that about 2015 the smartphone dethroned the PC as the primary driver of basically all consumer technology innovation.

Reducing bandwidth is a more pressing concern for today's congested cellular data networks. It's not as important for someone on a desktop PC sitting on a fat broadband connection.
Posted on Reply
#25
AddSub
cvaldesMy guess is that AV2 hardware decoding support will come to smartphone SoCs before mainstream consumer PC GPUs.

Remember that about 2015 the smartphone dethroned the PC as the primary driver of basically all consumer technology innovation.

Reducing bandwidth is a more pressing concern for today's congested cellular data networks. It's not as important for someone on a desktop PC sitting on a fat broadband connection.
Right, thats the point, this helps Google and their clients first and second and most likely 3rd, like I said. In the mean time we are measuring costs to clients in nuclear power plants, watts wise. LOL. YouTube has almost 3 billion users! Few watts times lets say.... 2.75bil users daily? Thats like lot of watts. A LOTTA WATTS!

Like I said, it's the paradigm of "NOT MY PROBLEM" engineering.

...
..
.
Posted on Reply
Add your own comment
Nov 13th, 2026 22:18 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts