2007/02/27

Cogentco ergo sum

A few years ago, I wrote paper entitled Cogent: Disruptive pricing or disruptive marketing? which showed that Cogent Communication's claim of revolutionizing the economics of Internet bandwidth was a clever marketing ploy, not a revolution in terms of "real" prices, costs or technology. It was just a lower quality network, and their prices were cheaper than the others, but not anywhere near as cheap Cogent made it sound. Nothing wrong with that... unless you don't like the "tragedy of the commons" effect.

Indeed, since then, instead of attracting their ideal customers (following what I called in the paper the "focus on retail" strategy -- get customers that pay $1000 for a 100Mbps and only use 1-2% of it), Cogent got much much more of the opposite -- "wholesale" customers that can really fill up the pipe, while still paying a flat $3,000 for a 100Mbps or $10,000 for 1Gbps pipe. And of course it's a self re-enforcing cycle, because the more that happens, the less attractive it becomes to any other kind of traffic. Now recall that 2-d table showing their profit margin as a function of utilization of the pipes, and transit/peer ratio of the traffic... Thus, basically from 2001 onward, they were going toward the bottom of that table (higher utilization), where it's mostly red. Which can't possibly go on forever right?

In principle it can't, unless they last long enough to move to the left of the table as well, i.e. reducing their transit/peering ratio. In other words, the other solution mentioned in the paper, the "buy their way into Tier-1 status" approach. Well they tried. Naturally the incumbent Tier-1s had no desire to admit them into the club, anymore than Rome would've opened the gates for barbarians. But would it be possible for Cogent to force it's way in? The only way is if you have a large enough customer base that you can go to Level 3, AT&T, Sprint & co and argue: "Hey you need to get to my customers as much as I need to get to yours, so let's be peers." Once you've done that with each one of them, then you are a Tier 1. In recent years only ATDN, the transit backbone of AOL has succesfully completed that series of moves.

Unfortunately for Cogent's hopes of muscling into the Tier-1 club, most of their customers were wholesale buyers that were multi-homed, i.e. had at least one other transit provider. Which is just common sense from the buyer's point of view, you want to shovel as much traffic as possible into the cheap Cogent pipe that has a flat rate, but for sensitive or "premium" customer traffic, you want the option to route through another better quality transit provider where you pay by usage. Given that situation, theoretically if Big Transit Provider X decides to "dis-peer" its network from Cogent's, the main thing that will happen is that Cogent customers will be pissed at Cogent for making them use their expensive other transit to get to those destinations (and to Cogent customers it appears like X is still "up" but Cogent is not giving them full Internet transit).

So guess what ATDN did when they were safely in the Tier-1 club? In Dec 2002, they dis-peered Cogent! (Its one of those clubs where you have to close the door behind you). Meanwhile those red numbers were piling up for our protagonist. By 2003 it was on the verge of going bankrupt and eventually it essentially did go bankrupt (I'm no accountant but I think what happened was they couldn't pay for their vendor-financed routers from Cisco so they gave Cisco a whole bunch of equity instead -- if you have any friends who owned pre-2003 Cogent equity, ask them... Then again, maybe you shouldn't remind them). But even as the company ha done foot in the grave, they kept selling those GigE pipes at $10,000/month, and it was still the cheapest bandwidth around, so their traffic volume continued to grow by leaps and bounds, making them one of the largest backbones in terms of traffic carried. By 2005, Cogent had grown up and was really pushing to the left, on that table. But Level 3 was threatening to de-peer (as was revealed later, negotiations had started in July) and they were big enough that that would push Cogent far over to the right. In an attempt to get more leverage against Level 3, in August 2005, Cogent reps started contacting Level 3 customers and offering them insanely low prices, even lower than their traditional "disruptive pricing", but only for Level 3 customers. Finally, in October, Level 3 called their bluff and before you knew it, Cogent was down, and there was no doubt as to who was Tier-1 and who wasn't. For more on the tactical details of this particular incident here's my opinion on the Level 3/Cogent peering disput of 2005.

So what now? Cogent is still losing money. But, there's one important variable working in their favor. Bandwidth prices have been declining more or less steadily at 30% decline per year for the last 3 decades. So now, typical transit prices went from $150-$200/Mbps/month in 2001 are now in the $15-30 range. So the good news for Cogent is that they may finally start making a profit on bandwidth! But the bad news is the marketing advantage of their "disruptive pricing" has shrunk dramatically. Today Cogent is just another transit provider, a bit cheaper and with a lingering reputation for lower quality among large backbones..

All the money they lost in the past by being cheaper does not seem buy them any competitive advantage today. It simply was a giant exercise in transforming the savings of little old ladies (the shareholders) into sports cars for porn webmasters (the heavy users who managed to get wholesale bandwidth below the true cost for a few years). Oh and Cisco ended up owning one of the major competitors of all their other main customers... Oops!

I think the moral of the story is that IP transit is a commodity, you can compete based on price and quality (or you can bundle it with other services and make it a non-wholesale product, like a CDN, hosting, managed services etc. but that's a whole other story). But there's no point in trying to build market share by selling below market (unlike the case with say software). When it's a commodity, sophisticated buyers (multi-homed, with their own ASN and IP address allocation, and connecting in carrier neutral facilities) will just arbitrage -- enjoy your early losses and leave you if you ever try reap the higher profits.