|
In 2005, UPC Cable's CTO estimated his bandwidth cost at 10 cents/gigabyte. That's less than $1/per at the three gigabytes Comcast and most other U.S. carriers estimate. That's 1-3% of the $30-50 price customers pay.
Here's the original DSL Prime news break from 2005, a figure since confirmed by other carriers. Three years later, an enormous amount of "advocacy" has convinced many that growing video will choke the web, prices should go up, and cable companies must throttle their customers. That's partly true at some smaller and rural carriers, who pay much more for bandwidth. It's simply not true at any of the large broadband carriers that are 90+% of the U.S. market. The exaflood is coming, but Moore's Law is bringing down prices about as fast. Odlyzko has pretty conclusively shown that Internet traffic per user has been growing at about 40% per year since 2002. Including the new users, that's 50-60% overall. The early data for 2008 suggests a slight drop in the growth rate. From DSL Prime three years ago.
Internet bandwidth isn't “free,” but it's become cheap enough to make DVD quality video practical while discrediting absurd bandwidth charges like BT's. Tony Werner, CTO of the largest cable company outside the U.S., has succeeded in peering 92% of his data traffic, dramatically bringing down the cost. During analyst day in beautiful Rose Hall, Werner noted that companies recently purchased by Liberty had been paying about 20 euro per megabit of transit. Because of his efficient volume buying, his direct cost is about 10 euro. John Malone's Liberty has major holdings in Japan, Holland, France, Chile, Romania and has just bid for Switzerland. On peered traffic, the cost drops to about 5 euro. I hear from others the U.S. is a little cheaper, smaller customers pay more, and much of Africa, Latin America, and India are high. That works out to $0.50 to $1.50 per customer per month, based on typical traffic patterns reported by other carriers, including occasional heavy downloaders.
About $1/customer/month is probably right for a large carrier in late 2005. I have been using a figure of “less than $2/customer” based on data now over a year old from several sources. More recently, I have heard bandwidth costs similar to Liberty from a mid-tier U.S. carrier, the remarkably efficient London Internet Exchange folk, and the buyer for a major video provider. I can infer similar costs from Randall Stephenson's ARPU and profit analysis, given his comment that one-third of SBC customers take a higher tier.
Translating the cost per constant megabit flow of traffic to a cost per gigabyte of data requires making assumptions about usage, but is important for analysis of video over the Internet costs as well as whether bandwidth caps and charging are reasonable. (They aren't. at least up to 50 gigabytes.) I have not been able to get firm data, so I am making assumptions that probably slightly overstate the costs. If you get four hours per day by 30 days per month, that is 120 hours per month. One-megabit continuous flow is about 500 megabytes of traffic per hour, or 60 gigabytes of traffic per month. For a company like Liberty, that's about $0.10 a gig at the rates Werner mentioned. BT, DT, etc. should do at least as well, especially given their existing worldwide backbones and share of major undersea fiber. A good confirmation of this comes from the carrier commitments to IP TV carried across their network. They wouldn't be possible without switching/fiber costs that are this low.
Traffic costs remain at or below typical marketing costs unless the user goes well over 50 gig consistently. Two gig or even 10 gig “caps” have no cost justification. They are partially a way to keep some users' prices high, and increasingly a tool to prevent watching video over the net in competition with the services BT and others offer themselves.
The costs calculated in this item are lower than now out-of-date estimates, projections from carriers trying to justify high rates, and the internal cost figures in some large telcos. Both Verizon and BT once overstated their marginal bandwidth costs and shadow pricing to their own DSL divisions. That raised profits in their (largely unregulated) internal backbone and produced “losses” in their DSL division, although the money stayed within the same company.
This should go down with Moore's Law, and fall by half in 2007 or 2008 and again by about 2010. The incremental cost of moving bits over existing fiber is largely the switching/routing cost, electronics constantly getting cheaper. In most of the developed world, there's enough unlit fiber for several more years.
|
