On YouTube’s Transit Costs

April 16th, 2009 by · 10 Comments

Over on his Communications blog, Brough Turner took a swing at Credit Suisse's analysis of YouTube last week.  Credit Suisse had said that YouTube's 2009 revenues would be about $240M whereas its costs would be $711M - which would be a spectacularly bad business to be in by anyone's measure.  However, they somehow came up with the number of $360M for third party bandwidth costs.  Brough's point is that this is ridiculous, as it discounts the common practice of peering to reduce costs.  He's correct about this, Google's scale gives them all sorts of bargaining power with internet backbones.  That financial firms don't really understand the economics of peering isn't surprising.  Heck, few people really do.  However, Credit Suisse should have known their number was ludicrous just by being the bean counters they are.

Why?  Because the IP transit market is really not very big.  Providers make money (or at least try to) via other services offered over IP networks, not via raw connectivity.   For a really rough estimate, just look at a few of the top IP networks according to various measures:  Level 3 Communications (NYSE:LVLT, news, filings),  Cogent Communications (NASDAQ:CCOI, news, filings), Savvis (news, filings) [a subsidiary of CenturyLink (NYSE:CTL, news, filings)] and TINet (formerly Tiscali).   The first two have in the past claimed to be amongst YouTube's main providers.  Level 3's entire high-speed IP revenues are known to be in the neighborhood of $150M/year.  Cogent's datacenter-centric business is only around $100M/year and that includes colocation and other revenues.  Before lumping it with other data revenues Savvis used to report 'other network services' revenues of about $100M but that included private lines.  Likewise, TINet's 2008 revenues were less than $50M.  If one extrapolates those numbers on the back of an envelope to the rest of the field, the top 10 IP transit networks generate less than $1B from that service and Google isn't likely to be buying much from anyone else.

Now YouTube generates vast amounts of traffic.  It is Big.  Very big.  But it isn't that big.  If they're paying $360M for IP transit then I want to know who's getting paid, because the money isn't coming out the other end of the tube.  Besides, the whole point of Google's vast array of datacenters and related infrastructure has been to control such costs.  You have to give them some credit for a few brain cells here and there, even if you are writing a hatchet job.

If you haven't already, please take our Reader Survey! Just 3 questions to help us better understand who is reading Telecom Ramblings so we can serve you better!

Categories: Internet Backbones · Internet Traffic

Join the Discussion!

10 Comments So Far


  • vijay gill says:

    Level 3’s entire high-speed IP revenues are known to be in the neighborhood of $150M/year.

    Citation please? Doesn’t seem to square up with what I heard.

    /vijay

  • Rob Powell says:

    It was in a conference call a couple years ago, I’ll try to track it down. I believe they have also mentioned in passing more recently regarding pricing pressure on high speed IP only affecting around 4% of their revenue. It could have gone up since then, I welcome any corrections.

    Level 3 makes most of its money off of voice (both wholesale and core), private line and waves, and enterprise services. Subtract those, then CDN services, Vyvx, colocation, and all the other data services like ethernet, vpn, etc – and you wind up with not much left.

    What number have you heard?

  • The_Highwayman says:

    Rob,
    Many folks assume LVLT to be one of the largest IP transit providers based on the rhetoric crowe puts out about so many packets, etc they handle per month…

    This is a bit misleading, because they count VoIP packets, CDN packets, etc, IOW they count anything that touches a router interface in the network as IP.

    I do not know the exact percentage, but a majority of LVLT’s IP traffic is settlement free peering.

    LVLT’s IP services are still not generating much revs and never really have, they tend to beat their chest about their IP services, and anybody that has been around long enough, knows that much of that IP traffic is peering.

    LVLT also has very few paid private peers.

    Just because the traffic touches a router interface means that it is pure IP router to router generated traffic.

    LVLT plays semantics and always have, they like to think they are God’s gift to IP, when clearly others were before them.

    Besides ,after the shopping spree, LVLT more closely resembled it’s competitors and imho actually became what they stated they never needed to become to dominate the telecom industry.

    I think your numbers are closer to reality than many really want to see.

    This can be directly attributed to LVLT’s penchant for stating things are always better than they always tend to be, and the fact crowe could sell you water and make you believe it’s moonshine.

    you know me pretty well Rob and lets face it and I will say it again, if LVLT IP revs actually matched their rhetoric on their IP services they would have been generating GAAP profits back in 2002.

    Another point that illustrates this perfectly is their gross margin story.

    Man, if they truly had GM’s across all product lines like they try to say they do, GAAP profit would have been achieved 1 year after the network was completed 🙂

    As I have also always stated for years, with these guys you have to continually watch what they do and not what they say.

    • craigp says:

      Most people assume they’re one of the largest based on the several third party rating systems that show them 1 or 2 based on various metrics. While each system has strengths and weaknesses, there’s no doubt that there is significant connectivity and traffic there, CDN or not, to make Level 3 one of the largest.

  • Rob Powell says:

    Highwayman, I don’t actually have any problem with how LVLT counts its traffic or where its hi-speed IP revs are. That ip transit itself is an ugly business and has been for a long time is self evident.

    My point was only that analysts trying to measure the costs of people who buy bandwidth need to keep the other end of the pipeline in mind as a sanity check for their numbers.

  • The_Highwayman says:

    agreed

  • mark says:

    Highwayman, your previous comment doesn’t make any sense at all.

    You said “but a majority of LVLT’s IP traffic is settlement free peering” and tried to use that to mean that they don’t handle much IP traffic. Of course most (or ALL) of their traffic is settlement-free peering. That’s because they are a “Tier 1” provider! That is the exact definition of a Tier 1 provider; that they do not pay someone else to take their traffic.

    I know this is way off subject from the original post, but of course CDN packets and VoIP packets should be counted as IP packets. Isn’t that exactly what they are? i think VoIP stands for voice over *IP*.

    • craigp says:

      Good points Mark. Whether the bits flow from onnet eyeballs to offnet content or vice versa, there’s a decent chance they’re billable. Bits on a transit free network don’t flow from one peer to another. The value is in having a nice balance of both increasing the network effect of the carrier.

  • Either way, YouTube is still losing money hand over fist. Which is an eye opener.

  • SDP says:

    Rob…great write-up. I agree with your conclusions that CS’ findings were flawed. All in all, there is no way that Youtube would have these kinds of costs for delivering their bits (even if you threw in all of Google’s other traffic). While the market is still in flux, it is true that the IP transit business isn’t that big.

    That being said, I do think that your argument missed some points regarding why this is the case. While the ISPs you listed are generally known as some of the largest in the community, they are obviously not the only ISPs…and in fact, you left out some of the biggest ISPs that play a critical role in delivering IP services (including for folks like Google). Obvious examples are Sprint, Verizon, and ATT…not to mention a number of global companies. Off-hand, S,V,T have IP divisions that dwarf the size of the companies you named.

Leave a Comment

You may Log In to post a comment, or fill in the form to post anonymously.





  • Ramblings’ Jobs

    Post a Job - Just $99/30days
  • Event Calendar