Not mentioned in the article but definitely the elephant in the room: The rise of TikTok streaming [1]
TikTok doesn't focus on gamers, but attention is finite. TikTok surpassed Twitch a while ago in revenue and it's accelerated since then. TikTok + Youtube (very profitable) makes Twitch at best a flat third place. Not a great business to be in when ZIRP ends.
Nine years after Amazon’s acquisition of the company, the business remains unprofitable, according to the people, who asked not to be identified discussing private information.
This ... is wild, I had no idea Twitch wasn't profitable. If you're pushing that much live video you really need to find a business model that works, especially when Youtube arguably does it better. Kind of surprised it took this long to start to wind it down.
[1] https://appfigures.com/resources/insights/20210924/amp?f=4
I've worked for a major streaming company and with several other major streaming companies (including Twitch); I'd like to think I have a pretty good understanding of streaming costs.
The "very" part is probably not true, and that's with Google's massive advertising empire behind it. There's a good reason why Google is going so hard at anti-adblocker initiatives.
You can't run a profitable video streaming service with "just throw the video on a CDN" as the strategy, you have to be the CDN because at the scale you're operating at, you'll be murdered by external CDN costs.
But you're also correct that Twitch's content makes it incredibly hard to serve cheaply. Same reason why Mixer (competitor by Microsoft) folded so quickly and Facebook Gaming and Youtube Streaming have been neglected projects since launch.
There's only two viable video streaming business models right now: your content is essentially free (TikTok/Snap/Instagram), or your bandwidth is essentially free (HBO before the divorce with AT&T, maybe Comcast-Peacock+). Netflix sorta had near free content for a while before the studios cut them off.
I think Amazon was betting that AWS efficiencies could push Twitch into profitability, guess not.
Several years ago I cobbled together all the pieces to make a live video streaming website [1] like Twitch.tv, et al. It's impossible to actually run at any kind of scale without being an ISP or Google-scale peer. Not to mention the enormous amount of DMCAs when people are trying to stream live sports, Pay-per-view, etc.
It was fun to play with it for a couple months. But there's no business here. Twitch.tv is also finding that out.
[1] https://github.com/benwilber/boltstream
What about all the pirate websites? There's hundreds of them all around the world, some supported with ads, others with subscriptions. How do they manage the bandwidth problems?
For them, content is free.
99% of the content is free for twitch and youtube too.
You’ve never heard of partnered twitch streamers or YouTube rev share?
They might pay creators a fraction of what they make on ads or other extras (digital bullshit like emotes), but none of that is in exchange for the content. The content is still being developed, written, produced, performed, and provided to the platforms entirely for free.
Unless there's some kind option where youtube or twitch signs a contract with select creators and then hands them huge sums of cash upfront for the costs of future content production, or to license the creator's existing content to distribute via youtube/twitch then the cost of getting that content is still just as free for youtube/twitch as it is for the pirate streamers that upload bluray rips.
Ad revenue is money gained from advertisers at the expense of viewers, it's entirely separate from the cost of the content being viewed.
For the record this exact thing does exist for bigger streamers. See Ludwig for example, he switched to YouTube exclusively because it was a better deal than his twitch offer.
I've seen paid exclusivity deals from youtube before (https://www.gamesindustry.biz/youtube-reportedly-paid-usd160...) and while I'm surprised they paid a streamer like Ludwig to defect I'm pretty sure those kinds of deals don't exist for 99% of the content that gets uploaded to the platform.
99% of page views? Absolutely. Mr Beast absolutely has a YouTube contract and absolutely drives percentage points of traffic.
That’s exactly how it works. And the fractions (on Twitch) are between 50% to 70%.
I still don't get how so many 123movies clones can keep popping up with seemingly every TV show and movie there, with probably nearly all users using ad-blockers and very few possible advertisers in the first place. They get taken down regularly then presumably deal with the effort of moving elsewhere.
They don't seem to be p2p either, they have some server. Maybe it's stolen infra.
Most don't host the content themselves. They iframe or whatever other hosts.
They embed other hosts, most of the websites usually use the same hosts but with different frontends slapped on top.
A lot of the pirate streaming services I see are either using torrents on the backend or they're abusing normal video hosting services and just replacing links that get taken down so that when the sites/apps themselves die they only need a new domain/app to point at the the same files on those same services instead of moving TBs of files around
Sometimes bandwidth too if they have clever peer-to-peer setups.
a) you don't waste money on the licensing in the first place
b) you find the hosting where you can just pay for bandwidth
c) the bandwidth is actually cheap, until you hit petabytes of traffic or dozens gb/s of BW
d) you don't scale for millions, you scale for hundreds, maybe thousands
EDIT:
e) most users of such sites aren't demanding for quality nor service, therefore you can compress a lot more and if you overloaded they would come to other sites (which could be operated by the same people just serving from the other place)
Actually, the opposite is true.
If you live a country where piracy is persecuted and these services aren't available and you want to watch Silo, you don't have a choice — you have to pay for Apple TV. If you want to watch Stranger Things, you have to pay for Netflix.
But if those services are available to you, then you would have those shows on any of dozens of pirate services, and you can switch between them anytime. At which point they actually start to compete on other qualities, like streaming.
In fact, many of them have much better quality, at least in the areas where their audience lives.
It's not 'opposite', it's some people care but most are don't. You can cater for those who care, but that means costs in services and people, or you can ignore all that noise and print money from Regular Joe, who bought 15-in-1 DVD mere a decade ago.
c) and d) don't make much sense
Bandwidth might be cheaper in absolute terms, but you're not paying less per user than if you scaled higher. How do you go from there to "the business model is to not scale up"
Except you need more money thrown in the hardware to be able to utilize that bandwidth. It doesn't scale linearly. At some point it's easier to setup another node to handle another 10Gbit and while you now pay double the node cost , now you can actually scale horizontally, because it's easier to serve the same content to different people from many nodes than from one big fat one.
Also, you need clients who would watch your content and would give you ad views and hence money. If you pay $1000/m more for another 10Gbit for each node in the first place then it's another $1000/node you need to get each month.
Remember, this is not enterprise customers, with $/user costs, this is profiteering from a highly fluctuating and risky market of solving the service problem... without paying the IP holders anything. Though that's a B2B service problem too, but it wouldn't be solved till corporations' greed would exist.
For them the viewer may actually be the product if they are distributing malware, mining coins in the background, or hijacking bandwidth. One reason to pay for a service is avoiding being a product. Rich people will pay out the nose for discretion because information about them has serious value.
I don't know about "Google scale", but you definitely need to be able to peer to get the costs under control. That requires a certain level of investment (on the order of millions of dollars in hardware/datacenter costs), plus the human relationships to get the peering agreements done. Twitch absolutely had that in place, so it's not really fair to compare it to a hobby project and conclude the business model doesn't work.
There's absolutely a business, if you have license to good content that people want to see. If you don't, then you're in a horrible commodity market -- you're reselling bandwidth.
I feel like free peering is a thing of the past. Verizon/Comcast killed it by forcing people to pay for peering relations about a decade ago.
Free peering always was just "Hey, I have 10GB of transit to push to you, and you have about the same, let's keep the beancounters out of this."
As consumer Internet grew, that became more and more unstable, until Netflix hit and just absolutely overwhelmed it.
In my experience it's typically a good idea to keep the bean counters out of things.
Recently buying computing hardware I selected an American provider for my eu based lab. After paying the invoices the bean counters then asked if I could get the money back because they didn't want to pay VAT. I said feel free to ask for the money back yourself I'm happy with the purchase that you already approved. They refused and kept going up the chain until the director said just pay the VAT. after countless emails, meetings, and other time wasting activities, VAT ended up being 8 euros. Bean counters create drama to justify their bean counting.
I'm sure there are good reasons for them. But they are often simply in the way.
Entire industries exist to bypass the bean counters. Arguably much of SaaS just is that.
But you can make a comfortable living as a small VAR by just buying product and charging more to companies for it in the way their bean counters like to see it.
$1000 item with $8 VAT line item? Unacceptable.
$2000 same item with no VAT? Perfectly cromulent.
Bored engineers tend to over-engineer and build overly clever systems. I guess bean counters get bored too.
Verizon/Comcast don't even exist here, and I live between India and China, almost half of humanity. Are you talking in hyper-local terms ?
Peering exist everywhere, and at scale you have to pay, yes.
How is Amazon not a Google-scale peer?
The bandwidth was obviously never the problem for Amazon. It was the actual content. How do you monetize Twitch streamers and, more importantly, Twitch stream viewers? They never figured that out, even at Amazon-scale.
Force people who want to sell on Amazon Marketplace to dedicate product to and pay for placement within Twitch streams. Then identify those products in the streams and spam the viewers & chat with purchase links and content streams that contain related recommended products.
Turn the whole thing into a self-referential product placement & affiliate scheme.
Depending on how in in-your-face that is, it might just kill the platform. I don't have any definitive answers but maybe if they ramped up their own game development studios, and flooded the ecosystem with them, they might see some return on investment (people paying for micro transactions basically) that seems like a semi-organic fit for the platform.
This seems like it was already part of the plan, see lost ark + new world launches and ongoing prime support.
I don't understand why multicast isn't used. I imagine starlink multicast will become a thing and maybe we will all get transceivers. Everything old will becomes new again...
Multicast on the public Internet has never been much of a thing, and seems unlikely to ever be much of a thing.
(Some of us have been dreaming about it becoming reality for decades, but some of us have also given up years ago.)
I’m the founder and cto of a niche streaming platform. You don’t need to be google scale. You do need a different business model.
It's Amazon, anything niche is just not registered...
I’m sure something like this already exists but can’t there be a distributed streaming platform that pays people fractions of a penny for hosting (all/part) of a video like bitorrent. I’m sure there’s some cryptocurrency equivalent. Or even better some combination of CDNs and distributed bitorrent like streaming. At the very least, it should atleast ease the burden of massive central hosting.
Why does “content is essentially free” apply to TikTok and not Twitch?
I don't really follow the space, so if I use the wrong amounts or titles, someone please correct me.
For streamers with even a moderate following, they'll be in some status with Twitch to earn money off Twitch's ads. Don't recall if it's Affiliate or Partner or what, but the follower counts were surprisingly small to qualify for ad revenue.
Twitch pays out $3.50 per 1k ad views to those creators, so Twitch is paying for their content.
Tiktok has something that pays out to creators, but it's $0.20-$0.40 per 1k views and you need 10k subscribers to qualify (this seems much harder than Twitch's, which is in the hundreds iirc). It also looks like it was a one-time thing, where they set $300 million aside for that but I haven't seen any plans to continually fund it.
$3.50 per ad view is a very different number than $.20-$.40 per video view.
I wasn’t aware of the “one-time” aspect of TikTok creator monetization, but I just find it hard to believe anyone is going to make good content without being paid for it.
The issue is you are thinking too narrowly by only looking at payments per view, ad, etc. Influencers don't care so much about that so much as the value of views as a metric. They don't get paid by the platform, they get payed by their audience and especially by advertisers for product reviews, placement, etc. But just like twitter, tiktok is full of people with no serious plans to get famous who put effort into making content. Same with HN. It's fun to shout creatively into the void when you get a bit of validation back in the form of conversations, arbitrary points, views, etc.
Doesn't this imply that Twitch needs to cut payouts then?
In the long run it seems like streaming as a career that is solely funded by the platform is dead. Streamers will need alternative sources of revenue (patreon, OF, etc) to keep things going.
That's already how it is among the SC2 twitch streamer community (and no doubt true for BW and other games with small but dedicated audiences).
Mid-tier streamers get the majority of their money from direct donations (which display a notification live on stream) and twitch 'subscriptions' the viewers buy for $5, split roughly 50/50 between streamer and twitch (usually to unlock access to their VoDs or other perks). Which really isn't the same as passive ad revenue from the platform, as users aren't exactly incentivized to give this money directly to twitch.
The successful ones actively solicit or otherwise encourage subs and donations, and those who struggle with this have generally moved on from the platform or away from streaming as a career. Many of them also have patreon for extra perks, plus a youtube channel with a more curated selection of videos (which I've heard pays better for the time investment, since the content is mostly already created for twitch streaming or sometimes other sources).
At this moment I'm watching a guy whose monetization strategy is a roulette wheel program with challenges and dumb stuff for him to do that gets rolled whenever a viewer donates above a certain threshold. As I am posting a viewer literally just hit "play 100 games" (obviously very low percentage roll, first time I've seen it) and the streamer is pleading with chat for them to let him play the 100 games over the weekend instead of right now. Pretty entertaining, and a funny coincidence to illustrate my point as I'm typing this.
Another guy who is a caster (like a sports announcer for e-sports tournaments) hosts tournaments via donations, ie every donation has a portion go to the prizepool and tournament costs. He has relationships with all the top pro-gamers in the scene and gets great line-ups for relatively small offered prizepools by being a community figure and also being a great organizer and scheduler, etc to accommodate the pro players to make it effortless for them to participate.
It's been this way for many years for streamers trying to make a living with 1-2k average viewers and below, typically working 40ish hours per week if I had to guess (not counting practice and any professional obligations with e-sports, sponsors, etc).
Anyway, I love professional SC2 even though I haven't played for years, and I spend way too much time on twitch.
Majority of TikTok's user base watches what TikTok decides to show them, which TikTok can optimize to lower their content creator payouts (and further optimized by only showing content that have been properly primed in the CDN). TikTok's monetization is also pretty nebulous and also terrible; Youtube's "bad" monetization rate is like $4 CPM, TikTok's is allegedly $0.001.
People choose what they want to watch on Twitch, which they have to pay as a portion of subscription costs to the channel.
I feel like you’ve never actually used twitch. What you’re describing in terms of how it works and “subscription costs” is very incorrect.
No, I've just worked with Twitch and had friends in executive positions at Twitch. Many of their top streamers that pay the bills for Twitch are recruited and have employment/partnership contracts. They're very much paying for content during the time I had connections with them (pre-Amazon and a while after).
The nobodies that no one watches doesn't cost or earn Twitch very much.
Your explanation of twitch monetization (or costs?) was hard to understand:
Who is "they" here? Twitch paying the streamer (cost for twitch) or viewers (aka subscribers) paying the streamer (revenue for twitch)?
Twitch pays streamers?
Twitch "pays" streamers with a cut of the money they made off them though. The paycheck is 1/2 a cut of money given to the streamer by a third party with no work from Twitch, and 1/2 a cut from ad revenues that were generated by the eyeballs accrued by the streamers.
Twitch streamers get ad revenue sharing, afaik TikTok users don't?
Because the top twitch streamers earn a decent living, whereas TikTok creators are paid (basically) nothing.
TikTok creators get paid as well (it's called TikTok Rewards)
Twitch pays out to creators 50-70% (some earn millions per year) TikTok pays essentially $0
I don't think these two categories are differentiated by if their content is free (despite they do have a deference in this regard), it's more about how the customers pay for it.
Hack, I'd say livestreaming (Twitch, the livestream part of YouTube/TikTok etc.) is a very different business altogether compared to normal "video streaming" (serving VODs).
Also when talking about livestreaming business, one can't ignore China. They have some massive platforms (in both cashflow and audience size) despite being unknown outside China. Like Bilibili, Douyu, Huya, etc.
I don't think any significant chunk of the Twitch infra runs on AWS.
Twitch seems to pay normal rates for AWS. And they are around the profit-zone for several years now it seems, just not making big money. So they are in fact very profitable, just not for Twitch, but for AWS and Amazon.
Twitch's differentiator is its backchannel, the viewer's 'chat'. Unfortunately for Twitch this only works if the whole loop is low latency. This makes it very hatd to optimize for cost. On top of this, it also makes adds even more intrusive and disruptive for the viewer.
One thing that surprised me is how they never managed to grow the VOD side. They basically handed that side, arguably a less costly and easier to monetize segment to Youtube.
For content producers, twitch often seems like a not the primary revenue stream. They put out lucrative sponsored content, or have twitch as an advertizing channel for their Youtube VOD/clips or their onlyfans. All revenue Twitch misses out on, while bearing the streaming cost.
How is it possible for them not to be profitable? Don't they take a 50% cut on subs?
Video is expensive, and streamers don’t like ads.
Amazon has for years been saying that they think that Twitch is under monetized. In their minds, 20% of watch time could be ads, just like regular TV. Streamers don’t like ads: it kills the vibe when your audience gets a 2 minute timeout. So streamers aren’t running enough ad breaks, and try to support themselves via memberships, merch and other alternative monetization methods. And for some, Twitch is only advertising for the real money-maker on another site. So Amazon doesn’t get the ad money they think they should get, and they only get a cut of memberships.
So then it’s a question of how many servers and engineers are needed to support Twitch, because that will determine profitability.
It’s too bad they are so unimaginative when it comes to that. I mean at the very least they should hand controls to the streamer to run an ad: right now it’s the algorithm that does it.
Also they should primarily look at other ways to monetize. Subs are still the best way, but not everyone has that kind of money.
One other big problem is that there former heads just didn’t get streamers and they ran a lot of big ones out of the building. One streamer with a crazy amount of paid members was treated poorly in an amateurish way.
I think they have new management now which at least seem to be a bit more in tune with what streamers needs are and some of the painpoints.
They do give streamers a button to run an ad break, but they force ads on you if you don't run them often enough. That is very common for many streamers, because the ad breaks hit a large fraction of your viewers and it sort of kills the conversation for everyone.
Is there a button? I recently became affiliate but haven’t seen one, there is just controls for setting the intervals and the amount of time, with more perks as you increase the timeslots.
I believe the "run an ad" button is for partners.
Ah I see, not there yet ;)
Can you use the /commercial command? That should fire off 30s of ads. You can also specify longer breaks with e.g. /commercial 120
Manually running enough ads should also stop the auto scheduler from running ads and disable prerolls too
Will try that, probably at the beginning of my stream when I have actually not started yet. Thanks!
I roughly remember that there was one for a time which allowed you to somewhat shift the scheduled hourly ads, but I think they might have removed it after testing. Through I think it also never worked really well either.
Twitch should have offered a merchandising business via Amazon. It feels like Twitch should have understood how their streamers make money, and ensure that all of that monetization happens via Amazon in some way shape or form, whether that’s a print-on-demand business or facilitating product placements.
Absolutely! That would be a great idea. They could leverage their know how to scan for product placement depending on category you are in. It would seem far more in line.
We never did find out why they banned doc.
I first bought YT premium after I found out the YT model of ad-free is ONLY the equivalent to buying Twitch turbo. A membership to the channel will still get you ads.
And having YT premium changed the game of how I consume all media.
Thanks, Twitch.
They do: as a streamer, you've got a button to trigger an ad break at any time. But the streamers rarely push it, for the reasons already mentioned. Video ads pay so little that it's not worth it to streamers to annoy their audience that way, when direct monetization methods which depend on happy viewers (subscriptions, shout-outs, merch) are much more profitable to them.
There's a real problem of incentive misalignment between the streamers and the platform re: monetizing the stream by sticking ads in it, and I don't think it can be resolved; hence the layoffs.
As far as I can tell most streamers most times are, or would be, fine with having roughly 3min of ADs per hour. If it's at an predictable time they can slightly shift around to match up with their content. It's a amount of time perfect to standup, stretch drink, something and similar healthy things you really should do every hour. (You can setup twitch that way.)
But that's just 5% of ads.
And if you change that to 10% it is, as far as I can tell, already in a area where most would seriously consider leaving the platform for good and at 15% I think hardly any streamer would still be on twitch.
Lets be honest the only reason normal TV got away with 20% is because it had no alternatives, and often anyway just ran in the background.
But most funny because how few companies buy ADs on Twitch you might just end up seeing the same 3 ADs in a loop for 20 minuts. As far as I can tell at least outside of the US twitch is sometimes even incapable to run a 3-5min AD break properly due to the lack of bought ADs...
Streamers can pick when to run ads - it's a simple command.
All of you talking about how streamers just can't pick when to run an ad have no idea what you're talking about.
They can schedule a few 15 second ads and then run longer ads manually to fulfill their 50-50 revenue share / preroll exemption.
It depends on the mode. They can choose the time, but then get awful poor ad-revenue. Or they take the juicy contract, and are forced to let twitch decide when ads are rolling. And with gaming content, even choosing the time is not always flowing well with your content.
Twitch gets 50% of subscriptions. "Bits" go to the streamer, 100% unless they used a third party app.
Streamers have to run a certain amount of ads per hour or they both lose a larger "share" of the ad revenue, and there is a preroll. Prerolls cause a huge impact in viewership so streamers don't want it.
The problem Twitch faces is that it now has multiple competitors: tiktok, youtube, and kick.
It's not just servers/engineers. Twitch has a lot of moderation issues and that requires a lot of labor. That's partly why the others are eating their lunch - little moderation.
Live streaming 1080p 60fps video to a channel with even just a few hundred viewers is a staggering amount of data that you need to push out. Multiply that by thousands of simultaneous streams, and the numbers don't look good. Because even though "Amazon" might be able to afford to send out all that data, "a subdivision owned by Amazon" still has to balance the budget and show that it can cover the cost for its slice of the pie... and Twitch simply can't.
I've been in streams with less than 100 viewers and seen hundreds of dollars in bits & gifted subs being spent in a single hype train. Are those really unprofitable for Twitch?
When someone buys a sub ($5-25 depending on tier), Twitch gets half the money.
Bits are a little more complicated. Donators have to buy bits ahead of time and then donate them. 100 bits is worth $1 to a streamer. But 100 bits will cost between $1.18 and $1.26 depending on how many bits you buy at once. Buying a larger block means paying a little less. (Side note: Once the streamer has received the bits, they can't be taken back, even if the credit card charge that bought the bits is reversed due to a fraud claim)
If a streamer with <100 viewers was getting hundreds of dollars in a single hype train, then either that streamer is very lucky and has a rich viewer that doesn't mind throwing around money, or I'd suspect something fishy is going on. ie, bits being bought with stolen credit cards, or the whole thing is a money laundering scheme.
It would occur regularly to several different streamers I used to watch. Double digit level hype trains were the norm and according to the leaked earnings they were making well over 100K on twitch.
I have often wondered when the big spenders are coming from export controlled countries, do the Federals ever get involved?
Sub prices also depend an your country. In some countries subs are noticeable cheaper then $5. Through I guess most viewers are in the countries where it's somewhat around that price.
How long was the stream? Doesn't sound unprofitable if it's 30min to 1hr (my uneducated guess), but then again idk how many streams get donations like that.
Most "normal" streams get their maximum viewer count around 2 hours into the stream. While streamers tend to have something like a waiting room for the first 10-30min of their stream.
A typical streamer which can life from it but isn't getting rich (or even wealthy!) tends to stream 6+h streams at at least 3 days a week more likely 4-5. Which additional time costs for stuff like filling taxes preparing streams, updating software etc. it's not too rare for streamers to have a 50+ hour week, without getting rich or even wealthy from it. While some pop of and get wealthy or rich it's not the norm. Most do so because they love what they are doing and/or would have problems with other jobs.
I have seen multiple cases of small streamers using Twitch as a form a therapy to help them to overcome social awkwardness or some kinds of anxiety. Also some cases of depressive or otherwise sick people using it to have something like a job even through they are to sick to get any normal job even if they didn't need it for money (to avoid brain rot of being stuck at home and maybe some additional semi-social contacts).
Naturally there are exceptions e.g. of streamers "capturing some whales" and making quite good money with a fraction of the effort of the normal case, but in the end this are exceptions.
I don't know what the streams cost Twitch, but that sounds like a lot of video. Sure viewers are hopping on and off and don't chew bandwidth while off, but still, I suspect 15min of someone streaming Twitch is a lot less valuable than 15min of watching YouTube.
Most streams with less then 100 viewers I have seen have maybe 1-2 gift subs when streaming 5 hours or so...
Through even on larger streams most gifts come from a relatively small number of people so there definitely are some outliers like you described but as far as I can tell they are the exception not the norm.
I thin calling it a subdevvision is already an overstatement I think it's not even that.
Amazone provides something like a "live streaming infrastructure service" which they sell to Twitch and others (including the competition like Kick).
For them buying Twitch was good as it allowed them to thoroughly test that service somewhat "in-house", have an initial customer for it to show what it can do etc.
But now that this is done for Amazone there is little value in Twitch, mainly:
- can it make profit (currently no)
- does it improve our image (surprisingly yes, a bit)
- does it allow us to sell more prime memberships (in the past for a short time yes, I don't think anymore)
- does it send bad signals if we shrink/close it now which could affect Amazone stock (somewhat, but if you slowly try to fix it show that you do and then if it doesn't work sell it probably no)
(parts in brackets are speculative)
Why do you keep calling it Amazone across multiple comments?
I which I would know I don't know how often I already ended up on amazone.de (a site selling farming equipment) or amazone.com (redirects to amazon.fr for some reason), it's quite annoying ;=)
Hollywood accounting. First off, remember they're paying themselves (AWS).
Large streams are more expensive but those get the subs. Ad revenues they are also splitting but they play 3 minutes of ads that interrupts the entire content every 10 minutes.
Just because Amazone owns Twitch doesn't mean they pay themself, they are still two distinct legal entities and Twitch is competing with other non Amazone owned companies (e.g. Kick) for the same resources. If Twitter would not pay Amazone I think it might even somewhat fall under competitive distortion/unfair competition.
Large steams do amortize cost while the bandwidth cost is somewhat scaling per viewer there are other cost which scale per used resolution and per region the stream is streamed to. So on a per-view bases large streams are often not the most expensive. But very small streams with viewers across continents have the highest per-view cost.
AD revenue is split but the main problem is Twitch ADs are not worth much and a bought way to little, often leading to no ADs playing at all for many (non US) regions.
They do not (typo?), it's depending on the streamers setup but ADs are normally more in the 3min per hour basis which is fine to be honest. There are sometimes some additional banner ads but they don't cover the content, don't "play" and are not that disruptive. And there is Twitch Turbo which disables all ADs but streamers still get the AD money. Through that isn't worth it's money for a lot of people.
It’s “unfair competition” for a company to use its own servers to host its own business?
More like Google sweetening the pot for Play store with ad sale discounts, less like Microsoft having the audacity to bundle IE with the OS. If AWS was Twitch Web Services it would be OK. But Amazon is not Twitch so it is anticompetitive.
So now you’re saying a company is not allowed to branch out and have any other applications on their own servers?
Should Microsoft also not be allowed to host online Office on Azure?
If Office and Azure are operate by two companies owned by Microsoft instead of MS directly then MS the one operating Office would have to pay the one operating Azure for the hosting.
AWS is not a separate company and even if they were, there is nothing stopping one company from negotiating with another company for special negotiated rates.
You don’t think Netflix is paying rack rates for AWS do you?
no AWS _is_ a different company then Twitch, Amazone owns Twitch but Twitch isn't Amazone.
and yes you can always negotiate rates but it has to be in some "reasonable boundaries" so Twitch still has to pay Amazone and Amazone can't just set the price to just the cost Amazone has, even less so below.
This is not in any way shape or form true.
When you go into the internal “phone tool” if you work for Amazon, AWS, Twitch etc you still eventually get to the same CEO.
(Former AWS employer)
This is not at all how it works.
Twitch giving money to Amazon is just a number in a computer somewhere.
Even if the government "forces" them to change those numbers on a computer, there is nothing stopping Amazon from just giving twitch more money to fund it, if they knew it were "profitable".
The real reason is that twitch buys AWS services at cost with zero profit to anyone, and even at those low prices it still can't make money.
Twitch is doomed.
How do you know Twitch is "buying" AWS servers so cheaply though?
Anyways, infrastructure costs is not as important as market share in the long run, bandwidth and transcoding costs will go down with time. If you build a platform where nobody has grounds to compete with you like Twitter, (Threads tried), you've got something really valuable - maybe not 44 billion worth but still.
I am supremely curious about this myself.
My guess would be their ad spend is bad in comparison to the other networks (Google, Facebook, TikTok). Could be because their audience isn't worth as much, could be because the ad formats are bad, could be because they can't get advertisers. But unless they are serving ads from Google's network, my guess is they are 4th place at best.
As noted on other threads, their streaming infrastructure must be a burning dumpster fire of money. Live streaming is super hard since you can't edge cache it, and it requires actual hard computer engineering to make work. If a mere mortal company were to try to run a clone of Twitch on AWS, it would run out of money in days. It also likely uses the same hardware as other high-value products, such as AI.
kick.com is trying to run a clone of twitch on aws for a while now. too soon to tell.
Running on VC money? AWS sounds like a terrible place to host your live streaming infrastructure unless cost is irrelevant
No they run on online casino money, the founders own Stake.com
I'm finding it curious that everybody here seems to be overlooking this.
---
(I think the thread is too deep and preventing me from replying directly to child post)
I posted here[0] a link to a video with pretty good commentary, speculating on and critiquing their strategy.
To paraphrase it, they are not necessarily banking on operational costs being dwarfed by casino revenue in and of itself... but that Kick will yield exposure and drive more people to their gambling platform overall. He is also skeptical and fairly unconvinced that it'll be a slam dunk venture for them.
[0] https://news.ycombinator.com/item?id=38935436
Thanks for the context - that is a great revenue source if morally dubious
I'm not an expert in this, but I don't think you'd be using similar hardware for video encoding and AI. Encoding uses a lot of compute, but it doesn't use much memory, whereas deep learning models (especially LLMs) use gobs of memory for both training and inference.
If I had to guess, I'd speculate they're doing encoding on their Graviton ARM CPUs: https://community.arm.com/arm-community-blogs/b/infrastructu...
yes encoders are special purpose hardware mostly unrelated to AI
for consumers they tend to overlap because they way to get a good encode is to buy a good graphics card which might be more expensive due to AI and before that crypto
through on server center there are special purpose encode PCIe cards I think, through the ARM CPUs might also be an option
as a side note because people mix it up all the time Twitch is owned by Amazone but a distinct legal entity which has to pay Amazone for any AWS/Server stuff they use in the same way as other customer
By spending too much on their execs
TL;DR: more operation cost then it seems, limited AD income, many streamer on which they lose money, increasingly more competition with "unfair advantage" (second platform effect, YT/TickTock; unethical content Kik(gambling); unethical monetization strategies TickTock).
operating a reasonable reliable live streaming service is quite expensive and Twitch is not Amazone. Sure it's owned by it but it still has to buy all computation resources from Amazone, for similar prices then some of the competition like Kick (which deeply interwinded with gambling streams which are profitable but ethically and legally problematic and hence banned on Twitch)
and while they do take a cut this is only profitable if a streamer gets enough subs/bits etc.
but most streamers on twitch do not get that, but might still have people watching across two continents+. E.g. having less then 50 viewers but viewers across 2 continents with a stream earning less 1€ income (not profit) per-hour in average isn't that rare
So the "big streamers" would need to subvention the cost of Twitch being free for streamers, which isn't easy. That some people do use Twitch as a form of marketing platform to goat people on other platforms and then there earn money from then doesn't help (and given how little twitch does against that but often simply could I'm seriously wondering if there is some internal employee corruption scandal to be uncovered).
Then there is another issue Twitch has compared to YT, it ADs are worth way less. Actually often Twitch doesn't even get enough ADs bought out so that some people in some regions sometimes see no or hardly any ADs (no one bought any) or ADs repeat too often...
Lastly and maybe for some people surprisingly given how some tech channels love to blow up any small negative news about twitch it's trying to act "reasonable ethically", i.e. not "oh they are grate people ethically" but "they have limits on how evil they act which are more then the law requires" ethically. You can see this if you compare the usage of dark patterns and highly addictive feedback loops between Twitch and TickTock. The later maxes out everything they can wrt. dark patterns which have addictive effects (at least for vulnerable people) and obfuscate how much money you spend. While Twitch, well does not do so.
Oh and probably there was some mismanagement, not now but a few years ago.
I wonder how much AWS gouges them on price
Indeed very wild. Ironic too, considering the co-founder (Michael Siebel) is a managing director of YC and has advised hundreds of startups about creating businesses.
I wouldn't mind being advised by someone who created a business and sold it to Amazon for a billion dollars, lack of profitability notwithstanding.
So they found the “greater fool”?
Mark Cuban also sold broadcast .com to Yahoo for 5.7 billion
That broadcast.com to Yahoo and Tom selling Myspace to Murdoch are 2 of my favorite flop acquisitions. Honorable mention: Digg selling its soul to their own sales team.
However, Geocities to Yahoo is one of my least fave.
I'd never heard of broadcast.com before, it is the inspiration for ROI (Radio on the Internet) and Russ Hanneman?
Tumblr to Yahoo for 1.1 billion in the hopes it becomes "The new PDF" was pretty good too.
I guess one could also argue in these cases that the founder started a very promising company, and the buyer failed to realize its potential.
You must be in marketing
All YC and VC-types care about is that the companies they either invest in or are associated with get a fat check from an acquisition or IPO. After that, they can fail for all they care.
He actually left Twitch pretty early on though.
I imagine live video is a lot more expensive too. Can't just throw that on a CDN. And unlike YT or TikTok videos with sort of a time limit, Twitch streams can run for hours, chewing up bandwidth while so many "viewers" are giving divided attention at best.
You'd be surprised. I have no idea what Twitch does now, but when it was JTV, we had quite a bit of our streaming video on CDNs. We built a complex infrastructure to arbitrage bandwidth.
But yeah, if you do it naïvely, you'll hemorrhage money.
Right, I figured you can place it on a CDN carefully, but it's not nearly as easy as with non-live video which I'd describe more as "throwing" it on there. And even with the complex infra already built, it must be still more expensive to operate. Everything you cache is only good for 10-15s.
It's more similar than you might think. The big problem is dealing with the bursty nature of live video -- lots of people will start watching a live event at the same time, which can easily trip you over into expensive bandwidth tier for the peak streams. You need to be able to react in real time and reallocate connections to cheaper providers, shape traffic, use peering arrangements, etc. Static video providers have the luxury of time, and can make these decisions slowly.
I forget the exact numbers, but it wasn't totally uncommon for early JTV to burn tens of thousands of dollars on a single stream before we got it under control. Someone starts a stream of a soccer game involving (say) Turkey, and it was like lighting money on fire. The whole event might be over before you could flag the thing down.
But like any other form of video, most streams get no viewers, and are relatively cheap to host when you have a scaled infrastructure.
Explains why they broke go-live notifications. It's meme-level timely now, sometimes 2-3 hours after they went live.
Hmm yeah, now that I think about it more, a 1hr stream with 10K viewers is no more bandwidth in theory than a 1hr static video viewed 10K times. But there's what you said about burstiness challenges, which is interesting to read about.
You can absolutely throw live video on a CDN at which point the bandwidth cost is the same as on demand video. The real-time encoding costs are the real issue.
The cache TTL on VOD is significantly higher, so your whole infrastructure needs to have a live origin, which gets hit repeatedly to get the latest fragments/segments from the origin, from multiple edges/midgresses. I find that our origin consumes an order of magnitude more CPU than transcoding and packaging for live once you have even a couple of thousand live viewers and when serving LL-HLS/ll-dash with a low fragment size it increases more. if you are latency sensitive then i between caching layers are not an option.
I watch 8-12 long videos on YouTube, but I get your point
I'm not surprised a bit given how hostile Twitch has become to users. Completely irrelevant ads, ads playing at worst times (who the hell thought it'd be a good idea to show an ad before the actual stream??), apps abandoned(at least apple tv app). I literally stopped watching some streamers because of Twitch itself.
This is in-line with Amazon's departure from "customer obsession" overall. I can't wait for bigger tournaments to move to a better platform.
I watch starcraft on twitch but can never watch it live. It'll get to a crucial moment and then a 3 minute commercial interrupts. Literally the game is over at that point. No point to watch.
They'd have done better with some sort of integrated commercials or banner bar or something else.
Or even just buffering those 3 minutes. So I do watch on twitch but I wait till the match is over.
What also sucks about twitch is you can't rewind. So many problems there
You can sub to the channel or get Twitch Turbo to remove ads. One sub is free if you have Amazon Prime.
If I have to pay then I might as well wait for the vod
If you load the VOD while they're still live, you can pause in place and rewind. YouTube does it natively in the live broadcast, but Twitch being Twitch just isn't incentivized to make it on the actual live stream, probably because money.
Twitch allows streamers quite a bit of control over when and how ads run, including manually triggering them. As long as they run a certain number of ads per hour, you don't see a preroll.
The streamer you were watching didn't bother to reduce ads to short 15 second chunks or run them manually during slow points like sitting in-queue.
Ads are controlled by the streamers. They can do prerolls if they desire, or you can at least control whether or not if they're running by rollings ads mid-stream.
And apple tv app isn't abandoned, I use it every day.
that being said, you're not compeltely off-base, some streamers are now simulcasting on youtube , presumable to mitigate some of this.
This is not entirely true, at least not anymore. Twitch forces ads every two hours? or so nowadays (some predefined amount of time), even if the streamer has ads disabled entirely. I believe it's either that or preroll ads? Don't quote me on this but I've seen it happen and explained like that by more than one streamer.
Prerolls in particular are super annoying to the point that I've moved on to twitch turbo, even though I have a subscription to the vast majority of the creators I watch regularly. Just opening a stream of some other creator to see quickly what is going on and just getting slammed with a 30s+ AD is not a good experience.
In fairness to twitch though, they do cover all the bandwidth usage from all these streamers, including the ones which don't even qualify for partnership and ad revenue. I'm just glad they offer the option to pay to get rid of these things.
Also not mentioned is the recent rise of Ad-Free ( yes completely AD-FREE as in not a single ad ever! ) platforms like KICK.com [1][2]
Amouranth used to stream exclusively on Twitch.com but now has moved mostly to Kick.com and streams the majority of her time there.
In this respect, Twitch's draconian content policing needs mentioning. People would receive week-long bans for unnamed reasons or super petty things. They rode the woke train as hard as they could and it was a sight to behold! They've recently slightly relaxed some of their disallowed content policies.
[1] Kick - Statistics & Facts https://www.statista.com/topics/11394/kick/#topicOverview
[2] https://www.similarweb.com/website/kick.com/
Kick has the same cost or more then Twitch (they use the exact same infrastructure under the hood,it's a service sold by Amazone to Twitch and others).
They saved some cost on building the UI etc. by well semi-stealing it from Twitch after a leak but that's just a saving bootstrapping cost not long term operation cost.
When it comes to general advertisement friendliness they are far worse then Twitch, i.e. where Twitch struggles to sell AD slots Kick probably wouldn't even bother.
Like many other start up like companies Kick is not currently profitable and lives from investor capital hoping to make it through.
Kick is deeply intertwined with gambling and many streamers on it frequently are in legal gray area when it comes to gambling law in the EU, they tried to somewhat get way from it but AFIK failed for now.
Kick has many streamers which are often seen as ethical problematic, like being associated with neo-Nazis, making it a no go for most advertisers. But also many streamers stay away from it for that reason and also (maybe more so) the gambling reason.
So AFIK Kick isn't likely to make it longer, except maybe as legally and ethically questionable gambling paltform.
You can't just say they are going to fail if we disclude the main reason why they are going to succeed!
There is a ton of money in gambling. That's why it's a very serious competitor.
They have a rare luxury of being able to tell advertisers to shove it, if they can make all their money doing gambling conversions.
the problem with gambling is its often not legal to advertise it too much, so many of the streams doing so either are in a legal gray area or outright illegal in many EU countries
worse a lot of gambling is deeply intertwined with organized crime in many countries
so if they as long as they have gambling they basically have no much chance to succeed in many areas which are not gambling on a wider scale and might outright bared from operating in some countries without 18+ age verification, too
Remains to be seen whether Kick can last long enough to get to profitability. It's certainly advertiser-poison, since what you call "draconian content policing" is closer to what advertisers want to see.
I doubt kick is making a profit either
Their ad is pushing their own Stake.com gambling site and having contracts with streamers to gamble on stream
I'm shocked it's not profitable. I've watched a person close to me spend $100/mo buying bits and get back nothing more than a few reaction emotes and jumpscares. Twitch's take on that is 50% isn't it? How on earth is it not profitable running a marketplace with those parameters and so many whales spending compulsively for a few seconds of attention from their favourite streamer?
I've always thought similarly about Uber. All they're doing is providing the platform and don't have to supply cars, drivers, etc. They basically take a cut on all rides worldwide and just have to supply a map solution. It seems to me they should be pretty damn profitable.
I think it is opposite of Uber. Content is free. There is hordes of people willing to do it for very little.
But actually doing the hard lifting of moving lot of bytes for users is the expensive thing. Ordering taxi and then updating location every few seconds is pretty light load. And there is not too many users doing it at one time.
What nobody tells you is that the cost of hosting all that content, and streaming it live with little lag, is far more expensive than many would guess. Netflix can have boxes with cached content very near you, and your ISP even likes it, as the most streamed content tends to be the same for many users. On Twitch, it's not so easy: You really want everyone to be seeing things with minimal lag, so they can spend all that money buying bits.
Other streaming services, which can use all of the tricks to save bandwidth and compute, have eye-watering AWS costs. Even with their high take rates, I bet there's a lot of streamers that, once accounting for their share of infra costs, end up losing twitch money.
Roughly spoken, there are 5-10 million active streamers, but only 5-10k who even make a significant amount of money. So even while most of them neither see a relevant number of viewers, or any viewer at all, they still create a huge baseline of costs just by existing and streaming.
Any interaction attempt with big streamers is worthless - like holding up a letter-sized note to someone driving past you at 80 mph.
The majority of Twitch's costs are going to be revenue to AWS; I wonder if they are still unprofitable if you include the AWS earnings on those services. Or in other words, if AWS hosts Twitch at cost, is it profitable then? If so, or if it's even close, the business would still have value to them. Maybe this layoff is intended to get there.
Twitch’s video infra is not on AWS. Only the non-video stuff is. Video and networking costs a lot.
Oh, that's surprising. It's in-house then? I can't imagine they'd be paying a non-AWS provider.
Yes, it was a complicated part of the acquisition. Twitch owns a lot of hardware and has it in colos all over the place. There are some control plane components in AWS but those are relatively small.
There is a difference in the target demographic though. Even if the most successful channels aren't about videogames, it is the target audience. That is different for TikTok and Youtube with heavy implication about how profitable advertising can be.
I haven’t looked at TikTok streaming but it would make sense if this has risen strongly. Twitch’s biggest category became “just chatting”. This was always a category ripe for being stolen by someone else on its own. If TikTok streaming is rising, I’m wondering if they have eaten into the just charting category growth.
Adding to that, I was watching Doug DeMuros round on YouTube the other day.
He noted that his views were down, and he likely suspected some of his audience had moved to TikTok.
Or something along those lines :-)
Doug is a massive Car Review YouTuber - maybe the biggest?
I can’t imagine seeking out the same content I get on YouTube but on TikTok.
But then it’s not content I guess, it’s attention. Is TikTok just stealing attention and will watch anything that served to them on the platform they happen to on that day.
Makes you wonder how much power personalities actually have :-/
Consider that "unprofitable" quote includes the fact that Twitch probably gets AWS compute at cost (Amazon subsidies Twitch server costs)