Hacker News new | ask | show | jobs
by bitmasher9 896 days ago
1. I think the infrastructure costs is probably more than you think. It’s probably the single biggest line item. They have to encode every video stream to multiple formats, and then deliver it globally.

2. The core product is pretty mature.

3. Elon raised the bar for how large of a staff cut you can make.

4. Most of their revenue is split with content creators.

5. They lost their CEO about a year ago. New guy has to make his mark.

6 comments

Twitch just announced a partnership with Nvidia at CES to allow streamers to encode multiple versions of their stream on their own PC and push them to Twitch. Previously, only partner/affiliate streams had transcoded streams at various levels of quality. That might cut back on the amount of transcoding Twitch has to do server side…

They also said they’ll be supporting AV1 and NVENC codecs soon which can help reduce bandwidth costs. They currently only support h264.

This is interesting, as I had an idea to do this when I was working on a competitor service during lockdown. Our idea was to use this to cut the cost on transcoded renditions, and for streamers that didn't have the ability to generate multiple renditions but still wanted them, we would offer it as a service in exchange for a higher cut of their revenue. That said, if anyone gets a benefit from lower quality renditions being distributed, it's Twitch.

If you think about it, the less bandwidth that Twitch consumes through their CDN network, the less money a stream costs per minute. A viewer still watches ads with the same CPM, a viewer still subscribes for the same amount per month per broadcaster, a viewer optionally pays the same for ad-free viewing (Turbo), and a viewer pays the same premium for bits, regardless of what quality they're watching at. So, interestingly enough, the economics of limiting transcoding are more about limited capacity and ensuring a positive cost per streamer. If a streamer does not become an affiliate or partnered broadcaster on the platform, the only revenue Twitch gets from the streamer is the ad money from its viewers. By virtue, they want as many casters on the platform as possible to become affiliates, as being able to get commissions from bit sales and subscriptions gives Twitch more revenue streams for the same content.

So, to your point, they definitely want to be able to support more efficient codecs that are not as patent encumbered as h.265 (which IIRC did have limited support in SE Asian markets), because at the end of the day, if streamers are generating the renditions for the ABR ladder, and ultimately, folks are using lower bitrates, then overall, this cuts costs for the same operational cost.

> They also said they’ll be supporting AV1 and NVENC codecs soon

You mean HEVC (x265)? I believe NVENC is just nvidia's hardware accelerated implementations of existing codecs.

Anyways, that sounds really interesting for both streamers who want more control over their streams' encoding, but obviously for Twitch, as it saves them compute power (though upping their intake bandwith?).

I wonder how feasible it would be for most streamers. Pushing one stream vs 3 or so is quite the difference.

I've always been entirely annoyed that there's no way for me to "encode what Google wants" for YouTube so it can be available directly, at least at highest bitrate.
>3. Elon raised the bar for how large of a staff cut you can make.

Did he? Their revenue is way down, and they've had a lot of downtime. Unlike Netflix, they make most of their money on ads, not subscriptions, so more downtime is more money lost.

They're also on a huge hiring spree right now. Go look at how many roles they're hiring for. I don't think they'd be hiring that much if everything was fine and dandy.

The loss of ad revenue isn’t a consequence of downtime or losing staff, and I wouldn’t agree they’ve had “a lot” of downtime. Most of the drop in ad revenue is a consequence of content management decisions that seem to be a matter of principle for the company’s new owner.

It’s strange to see hiring as an indication that a company is doing poorly; regardless, obviously Twitter is in a position where they’ve needed to eliminate much of their headcount and replace much of the rest.

> The loss of ad revenue isn’t a consequence of downtime or losing staff

Firing the Trust and Safety team had a huge impact on advertising - that's why big brands find Xitter so toxic.

It is certainly also true that Musk's... emanations haven't helped relations with advertisers, but it is the lack of moderation that freaks them out, and also by-the-by chased off a noticeable fraction of the user base.

And now there's the bots:

https://www.threads.net/search?q=Twitter&serp_type=tags&tag_...

> Firing the Trust and Safety team

Yes, technically, adopting a principled stance favoring freedom of expression implies laying off people whose job is to infringe upon freedom of expression, but it’s disingenuous to cast that as a staffing issue rather than a change in policy.

And I’m sure advertisers also really hate the fact that even the ads on the timeline can get community notes. But, well, I’ve seen what other platforms turn into under the foul influence of advertisers and I’m frankly not interested. If that means it has to be subsidized by an eccentric billionaire out of principle, so what? Other billionaires subsidize much more toxic outlets, the Washington Post for instance.

I'm not seeing his principled stance. In fact, it seems business as usual. He still bans content he doesn't agree with and doesn't follow through with anything he says.
I actually love the community notes on ads. Really good stuff.
> they've had a lot of downtime

Twitter is possibly the website most well known for downtime, and has been since literally the beginning. There's no real measure for whether it's true it's down more than usual lately, and sensationalist journalism and anecdotal reports from people with an axe to grind aren't really reliable.

The closest thing I can think of is looking at Google trends, which shows that "twitter down" is pretty consistent over the years:

https://trends.google.com/trends/explore?date=all&geo=US&q=t...

There's a spike in Nov 2022 when the layoffs happened, but again that's not indicative that it's especially bad - just that people were especially whiny about it or anticipating it would be bad

This is what I find ... interesting about the engineering layoffs - was Musk right? Was there a large part of the engineering function at Twitter that was just existing on some mythology, and was in fact just not needed? If so, are the FAANG companies (and others) the same? Legions of engineering roles just bullshit wastes of $200K salaries plus share options and free pizza?

I utterly hate Musk, so firmly hope he was wrong. Just wish Xitter went down more.

> interesting about the engineering layoffs - was Musk right?

Yes and no. Twitter was started before so much of the common “web scale” Open Source projects existed. Cassandra, Kafka, Spark, Kubernetes, etc didn’t exist yet.

So versions of the aforementioned and then some were completely done from scratch at twitter, who has continued to maintain those projects despite them not being the premier open source offering in each of those categories. They also never achieved the scale of some of the largest tech companies where it makes economic sense to have say, a custom database or queue system.

Finally, Twitter has been using GCP for some of their new projects in the last few years, but they still had old legacy systems and teams to maintain that software, so they couldn’t fully reap the rewards of the cloud.

So, if you rebuilt twitter today you could likely do it for a fraction of the engineers they had at peak, by leveraging either cloud or popular existing OSS solutions. But that’s from scratch, not porting millions of lines of legacy code.

What Musk should have done was migrate various systems, then wind down teams. That probably would have worked, but been expensive in the short term.

So I think they cut too much, and I also think they honestly thought more engineers would be up for “hardcore twitter”, instead of quitting.

Musk forgot that all his very loyal Tesla employees may have also factored in their 10x RSU appreciation. It’s a lot easier to put up with insane working hours when your next years stock grant is 500k or million or something. Doesn’t apply at twitter.

Tl;DR: not a Musk hater at all, but he fucked up at Twitter.

> don't think they'd be hiring that much if everything was fine and dandy.

Usually you dont hire if your business is not working or expected to grow

Why fire then rehire so soon then?
How does that work for equity? It must be difficult for them to explain this to new people, as where is the exit-condition for liquidity
The infrastructure is bought from Amazon, who in turn own them. Makes you wonder if it’s actually unprofitable, on the whole.
I think this is a fair question.
If its anything like my corp, Twitch only gets a small discount, if any at all. It's expensive regardless. Network bandwidth alone is probably in the $20+ million a month range.
Well, that's sort of the problem here right; yes at AWS retail you are paying $20 million a month for traffic, but the cost to AWS isn't anywhere near that, in fact how AWS charges for traffic doesn't match to how it incurs costs at all. And in general having desirable services like Twitch and of course being ginormously huge means people want to peer with you and you have more leverage in those peerings, making them cheaper.
One note on line item 1: Only Partners and sometimes affiliates get transcoding. Everyone else is just whatever they stream at.
> think the infrastructure costs is probably more than you think.

exactly. most people with good enough internet stream at 6 mbps CBR. A streamer with 1k viewers costs them several terabytes! egress for a quick 2 hour stream.

> Elon raised the bar for how large of a staff cut you can make.

Every member of staff fired now has a beef with the company, and being a social media company those staff will probably be big social media users with quite a lot of influence. Some may set up competitors or just badmouth you and your product.

Elon discovered this the hard way.

>... and being a social media company those staff will probably be big social media users with quite a lot of influence.

Genuinely curious to hear why you think this is the case. It's my anecdotal experience that the average social media company employee is just an average person like you and I, with an average number of followers.

Amazon already had a long reputation as the Hunger Games of s/w so their rep can't go down much.
This is an “inside a bubble” take.

The idea that software engineers at Twitter are influential on the platform is laughable.

How many of these accounts are software engineers? https://en.wikipedia.org/wiki/List_of_most-followed_Twitter_...

There’s more employees than just SWEs. And you don’t need to be generally famous, you need influence in key niches.

Twitter has historically hired people with large Twitter followings into roles that are more external. Some of them have been interviewed by the news media (eg The Verge). I’d be surprised if it didn’t have an impact on advertisers.