Earlier this month, AWS released its Amazon Interactive Video Service, which purports to embed live-streamed video directly into your own apps and websites. As AWS stated in its release announcement, a prepared statement for the press, and an exuberant tweet from their CEO, Amazon IVS “uses the same video technology Twitch uses for its live streaming service.”
One of the FAQs on their product marketing page in fact states:
What’s the key difference between Amazon Interactive Video Service and Twitch, and other video sharing platforms?
Amazon Interactive Video Service is a managed live streaming solution, which lets you build your own interactive live video applications or websites to develop a valuable relationship with your audiences with increased engagement. Twitch and other video sharing platforms are video aggregators, so you have to send your audiences to them. They manage the viewer for you instead of you retaining the audience connection and experience.
As I am a Cloud Economist, my immediate thought was this: “I bet that if Twitch actually used this, their AWS bill would be hilarious.”
Now, actually claiming to speak for Twitch’s financials is damaging if people misinterpret what I’m writing about, so I’m not going to do that. Instead, I’m going to talk about a hypothetical service called “Twetch” that just so happens to mirror Twitch’s metrics exactly.
We can pull Twitch’s—whoops, sorry about that. I mean, we can pull Twetch’s metrics from TwitchTracker.com for the month of June, which is sufficient for our needs.
We can also pull up Amazon IVS’s pricing model, which is predictably Amazonian. Pricing for video upload varies depending upon resolution: “Basic” channels are anything up to 1.5Mbps and 480p resolution, while anything above that is a “Standard” channel. Pricing per hour is $0.20 and $2.00 respectively.
Viewing is, of course, a hot mess, as it varies per geographic region the viewer is based in (good luck predicting THAT!). The first 10,000 hours are billed at one rate, while anything above those 10,000 hours is billed at a slightly discounted rate.
TwitchTracker shows that Twetch’s viewers watched 1,629,195,120 hours of video in June 2020, while its streamers pushed 7,158,730 hours of content to those viewers. Helpfully, that’s all of the metrics we need to gather.
How much does it cost to run Twetch?
Let’s run the numbers a few different ways.
We’ll start by being as flattering as possible to IVS and assume both that every stream is at a crappy resolution, and that every viewer is located in North America or Europe. These regions are tied for the lowest pricing tier and clock in at $0.0375 for the first 10,000 hours for standard channels, then drop to $0.035 for the rest.
These numbers show us that Twetch’s lowest possible publication costs are $1,431,746 for June, and their lowest possible cost to stream to customers is $57,021,854.20 for a total of $58,453,600.20 a month or an annual run rate of $701,443,202.40.
Remember; this is the LOWEST possible number using published IVS pricing!
Let’s try the other direction: What if every Twetch stream was in full HD and streamed to the most expensive region: South Korea (a nation famous for its lack of online video streaming culture).
At this point, the numbers change significantly upwards.
Now, the streamers are costing Twetch $14,317,460 a month, while the viewers are costing them $765,722,006.40. This annualizes to $9,360,473,596.80 a year—but we’re not quite done. After all, this is a Serious Social Network! You don’t deploy something like this without Enterprise Support in place! Given its sliding scale, we calculate it in Excel via this simple formula (assuming your raw AWS monthly cost lives in cell B35):
It helpfully tells us that the monthly cost to add in Enterprise Support for Twetch is $23.4 million, or a healthy $280,848,707.97 per year.
What this tells us is that the cost of running “Twetch” on top of Amazon IVS could reach $9,641,322,304.77 annually—or a bit over $9.6 billion a year going from June’s numbers.
That number is beyond comprehension. And just as a bit of additional context, $9.6 billion is slightly under 25% of AWS’s annual revenue. It’s also roughly 10x what Amazon paid to outright acquire Twitch.
It’s clear that IVS only makes economic sense up to a certain point; beyond that point, even aggressive discounting by AWS (and don’t delude yourself: nobody even approaching this kind of scale pays retail rates) doesn’t work unless the discounting tiers are into the lunatic zone (98% discounts on a service are, shall we say, “uncommon”).
Past a certain point, the right answer for live-streaming content is clear: Just do what Amazon did! Acquire a leader in the space, forget you own the service and outsource your own online summit to the lowest and thus most unscalable bidder, turn its technology into a service you offer to your own customers, set up a sarcastically expensive pricing model for it, and wait for customers who are better at content production than they are wrangling Microsoft Excel to fall into your cunning trap.
The larger point that’s often missed
Now that I’ve had my fun, let’s get serious here. It’s easy to look at something like IVS and say that its pricing model is ridiculous because it won’t scale to a global level before it becomes cost-prohibitive. But that analysis often belies an unfortunate truth.
The Last Week in AWS website, for example, runs on top of WordPress. On some level, analyses like the above are like saying that WordPress is a bad choice because Last Week in AWS won’t be able to handle WordPress’s limitations when it grows to be one of the top 500 sites on the internet.
That is exceedingly unlikely to happen. And if it does, there will be a series of architectural migrations long, long, long before it comes to that point.
In practice, customers are going to use IVS for an audience of a couple hundred people or so for a short period of time. For that use case, the pricing is reasonable and fair.
For the few that expand significantly beyond it, I would expect heavily discounted private pricing to be offered and/or those customers to rearchitect once they get to a point where they surpass IVS’ economic viability.
If this kind of analysis would be helpful to you, please reach out to our Microsoft Excel wranglers and see if we can help you predict your AWS bills before Amazon turns you into a mercy acquisition.