New research by analyst firm RampRate suggests that a previous report by Credit Suisse that claimed YouTube was losing over a million dollars a day was based on wrong assumptions. The number RampRate arrives at is way lower; they estimate a loss of $174.2 million a year.


The main difference is in the estimated cost of bandwidth; from the report:

Contrary to Credit Suisse’s estimates of a $470M annual loss, Google is more likely losing a fraction of that amount, due to peering for 73% of its traffic, buying bandwidth from some of the lowest-cost Tier 1 providers, using unprecedented bulk purchasing power to secure very favorable wholesale rates, and running data centers far away from expensive locales.

This assumption leads to a total bandwidth cost of “only” $75m instead of $360m in the Credit Suisse report. Brough Turner has already reported this earlier over at CircleID; there is just no way Google is paying for all its traffic. To quote from that excellent summary of the position Google holds on the transit market:

A similar effect plays out among Tier 1 providers. If one tier 1 network cuts a special deal with Google, Google routes all their traffic through this provider and suddenly the other tier 1 networks have large asymmetries in their tier 1 peering arrangements. Either they also cut deals with Google or they have to renegotiate their tier 1 peering arrangements to pay for the traffic asymmetry (something that’s highly unlikely!). Google is the one with leverage here!

What both analysts appear to be missing is an even more important development: Google has figured out that it has a unique position, and it appears to be building a “transit-free” IPv6 network, a situation in which they wouldn’t need to buy any internet bandwidth from any party. From their IPv6 FAQ:

To qualify for Google over IPv6, your network must have good IPv6 connectivity to Google. Multiple direct interconnections are preferred, but a direct peering with multiple backup routes through transit or multiple reliable transit connections may be acceptable.

So in order to get their users to Google’s services when they switch to IPv6, every ISP (currently) needs to get one or more direct connections to Google’s network; I’m sure the “multiple reliable transit connections” will only qualify if those transit providers have a direct peering with Google, so they don’t end up paying. In the US, they mainly rely on private connections to their network; in other markets they also participate in public peering fabrics. For example, in Europe, they have two 40 Gbps connections to the Amsterdam Internet Exchange, and are listed as having an “open peering policy”; in other words, they are willing to directly exchange traffic with any interested party.

google_amsixThey are also present at the London Internet Exchange; these are widely considered the two largest exchanges for internet traffic in the world. Notice that Microsoft is missing at the AMS-IX, and from what I’ve heard they are less open to peering requests. So it appears as though Google is working very hard to create a new competitive advantage that could save them hundreds of millions of dollars a year.

Google itself is keeping tight-lipped about their exact costs; their CFO Patrick Pichette has this to say about analysts:

“Most people build outside views of what it costs us to do things, and often they exaggerate,” Pichette said in an interview with the Canadian magazine Maclean’s shortly after Credit Suisse released its YouTube report.