Blog Post

YouTube Infrastructure Costs Vastly Overestimated: Report

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

YouTube is much closer to breaking even than widely thought, says a firm with intimate knowledge of global infrastructure costs. A widely publicized Credit Suisse report that said Google (s GOOG) would lose $470 million on the site this year neglected to account for factors such as peering traffic, wholesale bandwidth deals and cheap data center locations. Where the bank said YouTube’s costs will amount to $711 million in 2009, RampRate, a San Francisco-based company that advises large companies on IT infrastructure, says the actual cost is $415 million.

YouTube - RampRate vs Credit SuisseGiven Credit Suisse’s revenue estimate for YouTube, that would give the site an operating loss of $174 million this year. If you use other people’s revenue numbers — for instance, Jefferies said $500 million — the site would actually turn a profit.

Here’s where RampRate sees YouTube saving money.

  • Google uses peering for an estimated 73 percent of its bandwidth; Credit Suisse did not account for this at all. Peering is where ISPs trade traffic to reduce the cost of transit, and essentially means Google does not pay for that bandwidth. Google’s peering partners are known or thought to include Level3 (s LVLT) and AT&T (s T).
  • By negotiating for additional paid bandwidth, Google can probably bring down costs to 50 cents per Mbps.
  • From the report: “YouTube content is by definition ‘long tail,’ which means the storage it requires can be of a consumer-grade commodity quality. Google’s core competency is in managing commodity server/storage farms. At current rates, its servers likely cost no more than $500 apiece, and can add a sub-$100 1TB hard drive to produce a net cost of $.60/GB. Even with an aggressive 1-year refresh cycle, that is just 30% of the $2.35/GB projected by Credit Suisse.”
  • Google also lowers its serving and storage costs by using out-of-the-way locations for its server farms such as Iowa and Finland.

RampRate extends its analysis to say that maintaining an illusion of massive losses for YouTube is actually beneficial for Google. The theory is, Apple let the market think that its music-selling iTunes business was just a money-losing front to sell iPod hardware, so was able to keep its own costs and customer prices low, and therefore became a dominant force in music sales online. Likewise, the second YouTube is thought of as a money-making bonanza, its ability to negotiate goes down, and the lawsuits pile up even more than they already have. “I believe they’ve been very smart in managing the P&L to their advantage to negotiate better deals with creators,” said RampRate chairman Tony Greenberg.

Google also gets to use the tremendous YouTube audience and bandwidth to negotiate better deals for its overall business. “Basically their access to broadband eyes gives them an unparalleled negotiating chip to get lower costs,” said Greenberg. And further, through YouTube Google is able to broach new expansion initiatives like edge caching and co-location with the same ISPs.

From the report:

“Far from being an infrastructure money pit, YouTube is key to reducing
operational costs for other Google initiatives while also allowing Google to catch up to the superior
network performance of competitors like Microsoft (s MSFT), which currently boasts 10 times as many peers and 17% fewer hops to remote reaches of the Internet.”

Is RampRate right that Google’s non-monetization of YouTube is really a ploy? A scheme that conniving sounds a bit too much like a conspiracy theory. But we’ve seen via YouTube’s growth as a search engine how the site relates back to Google’s core business. Perhaps it fits neatly into the infrastructure expansion business as well.

19 Responses to “YouTube Infrastructure Costs Vastly Overestimated: Report”

  1. It is clear to me that YouTube can indeed negotiate free peering agreements. They are certainly building large POPs and buying some dedicated links for peak usage, where off peak can be used for standard free peering agreements with the carriers. I think this is a safe assumption.

    I am on the board of a large video sharing company and I can assume the following with confidence: I would be VERY suprised if YouTube were spending more than 50 cents per thousand videos streamed on their run-of-the-mill short videos, and the cost is more probably 25-30 cents. On a global CDN their commit rates probably allow them to cover Asian traffic “for free” etc leading to averaging down the overall cost. Assuming they are streaming indeed $1bn videos per day their bandwidth costs should be somewhere between $90M and $180M — which fits the above estimate. Very back of the envelope. The Ramprate estimate looks flattering to me but the CSFB analysts were clearly looking to grab headlines. Hardware opex also looks insanely cheap: I think CSFB’s guess is much better here and indeed there has to be some DB and so on. Transcoding falls under hardware kit and is dirt cheap.

    By contrast I would also be amazed if the company really cost $300M to run including content deals. If they have 300 folks, salary costs cannot be more than $45M which leaves a LOT of juice for content deals. On balance, $400M feels like a worst case estimate to run this business.

    If they did make $200M last year, you can hardly see them doubling this in 2009 (these kinds of volumes NEED direct selling if you want a decent rate card, even for YT / G). So $300M topline ?

    Probably still losing money but closing the gap fast.

    Once they are breakeven, the world can then quietly assess the full extent of what just happened to broadcasting !

  2. I have no idea how long google will be able to subsidize youtube users by paying for the infratructure. Google paid $1.65 b for youtube in Q4 2006. Looking back, it was a great exit for investors in youtube. This may turn out to be a bad investment for google investors.

  3. One other note, execs I’ve spoken to at YouTube say they are spending at least $500K a day on bandwidth/peering, which would put their yearly total at $182M, which is double the RampRate estimate and half of the Credit Suisse estimate.

  4. The numbers on all of this can be debated many ways as no one truly knows. But, talk to any Google exec off the record about YouTube and they will all tell you that YouTube is no where near close to breaking even in 2009. They are very clear about that.

    Also, while many want to debate YouTube’s bandwidth and hardware costs, what about their content licensing costs? Transcoding costs? Database costs? That’s something the Credit Suisse report included but I don’t know if this one does or not. There is more to YouTube costs than just bandwidth

    • Hi Dan, this report leaves all of Credit Suisse’s other estimates the same (see “content & overhead” in the chart), and just adjusts the infrastructure costs. As for Google execs’ comments, we feel confident saying that their infrastructure costs have been overestimated. But we wouldn’t publish comments that are off the record.

  5. Can someone explain why Level3 and AT&T would peer with Google for free? Peering is about the symmetric exchange of traffic volumes. There is no way Google’s traffic is anywhere close to being symmetrical.

    Maybe Google can construct a balanced load by pairing YouTube traffic with web crawling. But they Google would still fail the second part of a peering agreement which is for the connected networks to increase each other’s connectivity.

    There’s nothing in this for Level3 and AT&T so why would they do it? Instead they can charge Google for the bandwidth and collect a pile of money. There’s little Google can do but pay the bill.