I was wondering how much it costs Google every year to be running Youtube, and I decided to do some quick cost analysis.
This is what Youtube says about how much content is being put up on Youtube:
“60 hours of video are uploaded every minute, or one hour of video is uploaded to YouTube every second.”
OK, that’s a good start. And our reference video is 4 minutes.
youtube-dl -F (for listing all formats) shows me:
37 : mp4 [1080x1920] 22 : mp4 [720x1280] 45 : webm [720x1280] 35 : flv [480x854] 44 : webm [480x854] 34 : flv [360x640] 18 : mp4 [360x640] 43 : webm [360x640] 5 : flv [240x400]
youtube-dl –all-formats (for downloading all formats) shows me that the total size of all formats/files is 340 MB of storage – for a 4 minute video.
I’m going to be a little optimistic and say that not all videos have a HD format and come in far fewer size options (in fact, probably a minority have HD video), so let’s say its about half that: 160 MB for a four-minute video (to make my calculation a bit easier to follow.)
Total storage for 1 minute = 160 MB / 4 minutes = 40 MB/min
Total storage for 1 hour = 40 MB/min x 60 mins = 2400 MB = 2.4 GB
As per Youtube stats, 1 hour of video is being uploaded every second, so 2.4 GB of video is uploaded every second.
* 86400 seconds/day = 207.36 TB / day
* 365 days/year = 75.686 PB / year
So there you go. Approximately 76 PB of video data is stored in Youtube every year.
At the time of writing, 2 TB hard drives cost approximately $100 or less.
So 76 PB x $50 / TB = $3,784,320 on storage costs
= $3.8 million in storage costs
So there we go. Approximately $4 million in storage costs per year, using commodity hard drives. (I suspect the actual number could range widely, given the variable lengths of videos, formats, etc.)
Next, on to networking costs.
Again, from Youtube stats:
“Over 4 billion videos are viewed a day”
“Over 3 billion hours of video are watched each month on YouTube”
This is a bit more interesting – not all stored videos are served up, in fact, only one format of the video is selected and shown to the user.
Let me choose the Error:  shortcode requires URL to be set MPEG4 of the reference video, which is 40 MB in size. Also, let’s assume the user only watches 75% of the video or only that much is streamed over the network.
40 MB x 0.75 = 30 MB for 4 mins of video
30 MB / 4 mins * 60 min = 450 MB / hour
OK, over 3 billion hours of video are watched each month of Youtube. That is:
100 million hours a day = 450MB / hour * 100M = 45 PB/day
Per month = 45 PB * 30 = 1350 PB = 1.35 EB
Per year = 1.35 EB * 12 = 16.2 EB (exabytes!)
Wow. So 16.2 exabytes per year are sent through Youtube alone according to my calculations.
Now, I suspect I’m widely off mark here. For example, in Wikipedia’s entry on Exabytes, it says:
“Internet video will generate over 18 exabytes per month in 2013.”
And from this PC Magazine report, it looks like Youtube only accounts for 10% of Internet traffic. Note that that’s all of Internet traffic, so video is probably a great majority of that (if “HTTP”, according to the report, is 20%, then video is at 80%).
Still, I appear to be off by a factor, let’s say of 4x, so let’s say Youtube is only pushing 5 EB a year.
Amazon’s Cloudfront CDN charges $0.12 per GB for low-end customers. I suspect that Google/Youtube, through their combination of in-POP networks, working with ISPs, etc, is able to keep that cost down. Let’s say $0.01 per GB.
So $0.015 per GB = $15 per TB = $15,000 per PB = $15M per EB
Cost = 6 EB * $15M / PB = $90,000,000
= $90 million
So there we go. Approximately $90 million a year in networking costs. (Again, this number could vary wildly given the video sizes, network bandwidth required and most importantly aggressive CDN networking/technology solutions that I’m sure Google is deploying, as well as low CDN costs and any existing deals Youtube has with ISPs to reduce traffic for themselves and ISPs.)
So there we go, combining the two costs:
$4 million (storage) + $90 million (networking) = ~$100 million per year in networking+storage costs
This does not include the server count/costs for that, or personnel.
And I have to clarify again, that this is a very rough estimate, and I wouldn’t be surprised my numbers were off by a margin of 0.5x to 4x what I’ve come up with.