I can give some insight working for major streaming providers like Amazon Prime, BBC iPlayer and DAZN..
You gotta separate 2 type models - live and vod. Also 2 distribution models low concurrency with huge archive/channels versus high concurrency with small number of assets/channels.
Also 2 more variables funding( ads v subscription ) and coverage ( encoding profiles , DRM.and device support ).
Very fun field . I believe India cricket league achieved 100million concurrency recently and I deal regularly with 10,000,000 plus petabyte sized archives.
Looking forward to follow up articles. Reach out if you need any support
An interesting topic, and I look forward to reading the follow-ups. I feel like the subtitle is misleading; you don't actually explain why here, you just lay out the problem.
I think you actually touch on something interesting here, where there are social media like effects in video (why is X still popular after all of this), but many of the problems are fundamentally technical. You notably don't mention Instagram reels, which I suspect is profitable.
Yep, I'll be talking about Facebook video and Instagram Reels next week - part of the secret sauce for those Meta properties is that they're not single-purpose video sites. They're mixed media.
Beyond bandwidth a different challenge is encoding the videos to t he right codexes. People would typically upload a video in the highest resolution and then it needs to be encoded in the various other available resolutions. The video likely is also cut up and stored in smaller segments and that needs encoding too. Then there might be AI features like autogenerated subtitles or auto created chapters to consider as well and this is also something that should be done after the video is uploaded.
Crucially this stuff all needs to happen before anyone can view the video and has costs regardless if the video is a flop or is seen many times. It’s easy enough to move the video files to the right data center once it’s encoded but the encoding itself takes some work.
I would assume mobile first services can bypass this by not bothering with encoding to different resolutions since it’s mostly viewed on phone screens anyways (even if they have almost certainly studied what that means for retention). They might also be benefit from being able push more encoding work onto the phones (like when videos have subtitles on instagram) but that’ll depend on the specific service.
This is really interesting, and something that wasn't on my radar. It sounds like bandwidth is the real killer here, which makes me wonder how delivery costs will change over time. I assume cost/byte generally declines over time, but have no idea about the rate. Codecs also show efficiency improvements over time, but this rate seems small, maybe something like a 50% reduction in bandwidth per decade at the cost of much higher encode/decode compute. But consumers probably also expect higher video quality over time, which could easily swamp the above cost reductions.
Do we have detailed data on the cost of bandwidth/hosting/transcoding/etc. for any large video providers? I'm curious about the relative sizes but also the absolute magnitude of costs of serving video per hour/user.
For delivery to customer you can use rate of approx 1cent a gigabyte . Roughly most video can be streamed at about 4 cents an hour. These order of scale. But misses out 4 key costs
* Storage of archive at multiple locations for VOD
* Aquistion costs for live . That source to encoders/packagers
*Encoding costs for vod/live generally you need 8 video profiles across up to 3 formats
For long tail content your acquisition, encode, storage and packing costs can dwalf distribution.
Unless something revolutionary happens (e.g. 3D VR video becomes popular), I think we're hitting a wall in video quality demand. Most video is watched on phones, where the different between 4K and 8K is imperceptible.
To my knowledge, the human eye can't reliably tell the difference between 4K and anything above 4K. Even the difference between 1080p or 1440p and 4K is really pretty marginal imo.
4k makes a huge difference, when there is more than a dozen of dynamically changing numbers, icons, a stock market ticker, tiny men shooting projectiles the size of ants, whilst eight players scheme plot, scheme and backstab one another via color-coded text-chat that they don't properly understand half the time either.
In other words 4k is great for Age of Empires 2 team games viewed on a 40 inch+ screen and pretty much nothing else :)
I think if there is one company capable of challenging YouTube, it’s Spotify. The real issue is whether or not the math works out for paying creators and artists simultaneously. I do think they skittle at least try…
Well, that's simple enough. You can't compete with YouTube because they had first mover advantage then executed excellently for years, without any major mis-steps. Their product is free, makes people money and there just isn't any obvious angle on which competitors could gain traction other than higher creator rev shares, but even that isn't really obvious when you consider the huge amount of money YT spends on building audience.
I can give some insight working for major streaming providers like Amazon Prime, BBC iPlayer and DAZN..
You gotta separate 2 type models - live and vod. Also 2 distribution models low concurrency with huge archive/channels versus high concurrency with small number of assets/channels.
Also 2 more variables funding( ads v subscription ) and coverage ( encoding profiles , DRM.and device support ).
Very fun field . I believe India cricket league achieved 100million concurrency recently and I deal regularly with 10,000,000 plus petabyte sized archives.
Looking forward to follow up articles. Reach out if you need any support
An interesting topic, and I look forward to reading the follow-ups. I feel like the subtitle is misleading; you don't actually explain why here, you just lay out the problem.
> What are your options other than uploading to YouTube? Could you even name another site?
Many youtubers would say "Of course, have you tried Nebula? -- Now back to the topic of our video"
I think you actually touch on something interesting here, where there are social media like effects in video (why is X still popular after all of this), but many of the problems are fundamentally technical. You notably don't mention Instagram reels, which I suspect is profitable.
Yep, I'll be talking about Facebook video and Instagram Reels next week - part of the secret sauce for those Meta properties is that they're not single-purpose video sites. They're mixed media.
Beyond bandwidth a different challenge is encoding the videos to t he right codexes. People would typically upload a video in the highest resolution and then it needs to be encoded in the various other available resolutions. The video likely is also cut up and stored in smaller segments and that needs encoding too. Then there might be AI features like autogenerated subtitles or auto created chapters to consider as well and this is also something that should be done after the video is uploaded.
Crucially this stuff all needs to happen before anyone can view the video and has costs regardless if the video is a flop or is seen many times. It’s easy enough to move the video files to the right data center once it’s encoded but the encoding itself takes some work.
I would assume mobile first services can bypass this by not bothering with encoding to different resolutions since it’s mostly viewed on phone screens anyways (even if they have almost certainly studied what that means for retention). They might also be benefit from being able push more encoding work onto the phones (like when videos have subtitles on instagram) but that’ll depend on the specific service.
Not just right codec but following variables
* As you say closed caption generation luckily good cross standard platforms here
* Roughly you need at least 5 video 'profiles' to allow for bandwidth constraints on device
* 2 codecs at least .h264 but also .h265 and maybe vp9
* Packaging - this the container around the encoded fragments . 2 common standards are HLS and DASH
* Then drm . 3 common formats playready, fairplay and widevine. The cenc model is widely used but still means 2.
1 input and MANY outputs sadly .
This is really interesting, and something that wasn't on my radar. It sounds like bandwidth is the real killer here, which makes me wonder how delivery costs will change over time. I assume cost/byte generally declines over time, but have no idea about the rate. Codecs also show efficiency improvements over time, but this rate seems small, maybe something like a 50% reduction in bandwidth per decade at the cost of much higher encode/decode compute. But consumers probably also expect higher video quality over time, which could easily swamp the above cost reductions.
Do we have detailed data on the cost of bandwidth/hosting/transcoding/etc. for any large video providers? I'm curious about the relative sizes but also the absolute magnitude of costs of serving video per hour/user.
For delivery to customer you can use rate of approx 1cent a gigabyte . Roughly most video can be streamed at about 4 cents an hour. These order of scale. But misses out 4 key costs
* Storage of archive at multiple locations for VOD
* Aquistion costs for live . That source to encoders/packagers
*Encoding costs for vod/live generally you need 8 video profiles across up to 3 formats
For long tail content your acquisition, encode, storage and packing costs can dwalf distribution.
Thank you Mark, that bandwidth cost is actually less than I would have guessed. Makes sense that the other costs for long tail content dominate.
Unless something revolutionary happens (e.g. 3D VR video becomes popular), I think we're hitting a wall in video quality demand. Most video is watched on phones, where the different between 4K and 8K is imperceptible.
To my knowledge, the human eye can't reliably tell the difference between 4K and anything above 4K. Even the difference between 1080p or 1440p and 4K is really pretty marginal imo.
It depends on the content.
4k makes a huge difference, when there is more than a dozen of dynamically changing numbers, icons, a stock market ticker, tiny men shooting projectiles the size of ants, whilst eight players scheme plot, scheme and backstab one another via color-coded text-chat that they don't properly understand half the time either.
In other words 4k is great for Age of Empires 2 team games viewed on a 40 inch+ screen and pretty much nothing else :)
https://marknuyens.substack.com/p/spotify-videos My post expands on this idea, by the way. 😉
I think if there is one company capable of challenging YouTube, it’s Spotify. The real issue is whether or not the math works out for paying creators and artists simultaneously. I do think they skittle at least try…
I noticed you left out video services like Rumble. Was that intentional?
It may not be suitable for work, but OnlyFans definitely found a profitable video model…
Well, that's simple enough. You can't compete with YouTube because they had first mover advantage then executed excellently for years, without any major mis-steps. Their product is free, makes people money and there just isn't any obvious angle on which competitors could gain traction other than higher creator rev shares, but even that isn't really obvious when you consider the huge amount of money YT spends on building audience.