Cogent: Disruptive pricing or disruptive marketing?

Nemo Semret
Invisible Hand Networks, Inc.

"What if you could get 65 times more bandwidth than a T1 for the same price resulting in an Internet connection running at LAN speeds? How about eliminating issues such as quality of service, prioritization, oversubscription, hot-potato routing and caching simply by throwing unprecedented amounts of affordable bandwidth at these problems? Disruptive pricing as a sales tool and a competitive advantage is a simple, yet sound, proposition that will change the way businesses use the Internet." Dave Schaeffer, Founder and CEO, Cogent Communications

What if indeed? Cogent offers "retail" Internet transit service at $1000/month for 100Mbps, and "wholesale" at $3000/month for 100Mbps. That means 10$/Mbps/month, or 30$/Mbps/month. The most aggressive tier 1 ISPs sells transit at 150$/Mbps/month for large volume customers. Traditional tier 1s such as Uunet, Genuity, Sprint, AT&T, are well above $200. So how is it possible for Cogent to sell the same thing 5-10 times cheaper? Are they doing something fundamentally different, or are they simply burning money and destined for the same fate as Yipes, and others, who made similar claims and are now bankrupt? According to Cogent, the answer is that they have a newer, more efficient network than anyone else, with huge optical pipes, and that's it.

The full answer is more subtle than that, and it does not have much to do with "optical" -- all Internet backbones are made of fiber optic pipes. (Hint: Why is it that for almost anything else under the sun, "wholesale" is usually cheaper than "retail", while with Cogent it's the other way around?)

Cogent is not (yet) a "tier 1" ISP in the strict sense of the word (see definition below). They must get transit from some Tier 1s, and no matter how cost-effective their own network is, some of their customers' traffic must eventually go on paid transit to the traditional Tier 1s.

FACT#1: Cogent buys transit from Sprint and [Conversations with our Cogent account team in the Fall of 2001].

Of course, since they sell at a flat rate of $3000/month per 100Mbps circuit, profit depends on how much traffic the customer generates, and in particular on how much traffic is carried to its final destination via paid transit on a Tier 1 network.

FACT#2: Currently, 15% of Cogent's revenue goes to transit fees, 26% to other operating costs, and 59% is profit [Cogent presentation at Service Networks 2002]. This means out of $3000, $780 goes to operating their own network.

To achieve the claimed 59% profit margin, assuming transit costs are 150$/Mbps/month, for each 100Mbps customer, no more than 3Mbps of sustained usage transits via Tier 1s. This means either usage by customers is very low (which means that, on a usage basis, customers are paying a very high price), or Cogent manages to peer with enough other Tier 2 ISPs for free that very little traffic ever goes to Tier 1s (see table).

Profit per 100Mbps circuit sold at $3000/month

Transit traffic
20% 30% 40% 50% 60%70%
10%$1,920 $1,770 $1,620 $1,470 $1,320 $1,170
20% $1,620 $1,320 $1,020 $720 $420 $120
30% $1,320 $870 $420 ($30) ($480)($930)
40% $1,020 $420 ($180) ($780) ($1,380)($1,980)
50%$720($30)($780)($1,530) ($2,280) ($3,030)
60% $420 ($480) ($1,380) ($2,280) ($3,180) ($4,080)
70% $120 ($930) ($1,980) ($3,030) ($4,080) ($5,130)

The reality is that as of December 2001, the top 5 Tier 1 backbones carry more than 45% of total Internet traffic. [L. Roberts presentation at Service Networks 2002] Of course, for each individual source of traffic, the percentage that transits through the Tier 1s will vary; but in the aggregate traffic carried by an ISP, the distribution of all possible destinations is so diverse that it is safe to assume that it closely mirrors the overall distribution. Even if only 40% of Cogent's traffic goes on transit links, we conclude that Cogent's current profit margin is achieved because customers on average use less than 10% of the circuit. This would imply that the actual price per Mbps sustained paid by Cogent's customers is on average more than 300$/Mbps/month.

So Cogent has not changed the basic laws of economics. "Disruptive pricing" is nothing more than clever marketing.

Buyer strategy

Of course, as anyone who has been to an "all you can eat" buffet attest, with a flat rate pricing scheme, buyers with light utilization are basically subsidizing the heavy users. So, a smart buyer with lots of traffic can get a 100Mbps from Cogent, and simply use it more fully, pushing their effective price per Mbps much lower.

First, it turns out that is not so straightforward either. Because almost all the traffic eventually goes on to another provider (through transit or peering), the actual traffic that a customer can push is limited not by the bandwidth in the core of Cogent's own network, but by the connections to other networks. For example, our own tests in May 2002 show that no more than 40Mbps can be pushed on a Cogent circuit to a final destination on Level 3. Above that level, congestion occurs and quality degrades rapidly (latency rises to over 100ms, and packet losses become significant). This means that at the interconnection between Cogent and Level 3, there must be limited physical capacity or traffic management rate-limits in order to manager costs. Either way, in this case, the lowest effective price that we can achieve on Cogent is $75/Mbps/month ($3000/40Mbps), and even that is reached with degrading quality.

Chapter 11 for Cogent?

Second, as a direct consequence, Cogent will tend to attract more heavy wholesale users, while light users will get lower total costs and better quality by sticking to Tier 1s with usage-based pricing. As this occurs, not only will overall quality go down, but also, as can be seen from the above table, Cogent will be barely breaking even at best, and probably losing money on each Mbps carried. As volume increases, either they will go out of business or be forced to raise prices.

For Cogent, short of changing their pricing, there are two potential escape from this quandary:

  1. Buy their way into Tier 1 status

    Cogent has recently acquired the networks of Netrail, and PSInet. By buying other ISPs, they inherit peering relationships which help reduce the share of their traffic that goes over paid transit. However, this can only delay the day of reckoning, since by their own logic, these older networks cannot have the lower costs of operations that Cogent claims for its own new network, especially when you consider that integrating operations is going to be a long and expensive process.

  2. Focus on retail

    For small business customers, looking at this service as a replacement for T1 internet access lines, usually costing near $1000 for just 1.5Mbps, Cogent's $1000 retail price for 100Mbps is reasonable, but not cheaper. If they end up pushing more than a couple of Mbps, Cogent will simply reclassify them as wholesale and charge them $3000 (and by their own admission, they have done that in the past). But it should be noted that other operating costs, especially "last-mile" build-outs are much higher for retail than for wholesale customers which typically connect at network access points that Cogent is already present in.

In short, both of these strategies will result in raising operating costs, and so neither is likely to work as a strategy to presever their current pricing scheme. It seems clear that, if Cogent is to survive, sooner or later, their prices will have to come closer to other backbones' prices.

I bet that within 12 months (by mid-2003), Cogent's price for a 100Mbps wholesale circuit will be near $5,000/month, which will bring it on the low end of the range of transit prices, as other providers' prices will probably continue to decline at around the historical rate of 30% per year. The lower the quality Cogent's customers are willing to tolerate (congested peering and transit links), the lower the price can be. Of course, price increases can take many forms: explicit rate hikes; traffic limiting policies; a switch to usage-based pricing; blending of services inhereted from acquisitions; etc. Most likely, it will be a clever mixture of several of the above. For example, the current pricing plans could still be there, but with enough restrictions added that most customers will have to sign-up for the "new" more expensive service offerings.

Remember "free Internet access" providers? The most succesful one, NetZero, was built in a short time with brilliant marketing ("defenders of the free world" etc.). Now, NetZero is called United Online, and marketing itself as a lower-cost (but paid) alternative to AOL. The free service is limited to 10 hours per month. Replace "monetizing eyeballs with banner ads" by "terabit class routers with DWDM solutions and Internet-optimized architecture" and it's the same story. Revolutionary sounding jargon, but behind it, they will lose money, and the only way it can work is a) if they build enough market share before money runs out and, b) they can transition to higher prices without losing customers. It's risky (NetZero is the only survivor from the dozens of free access providers).

Conclusion: buyer beware

In the meantime, buyers should be aware that there is no such thing in the year 2002 as a Mbps for 10 or 30$/month. By the time you use it, it costs 75-300$/Mbps/month depending on your traffic (also Cogent sets BGP parameters that minimize your usage by subtly telling your router to prefer your other provider), the quality is pretty low, and the price will almost surely rise to about 125-500$/Mbps/month within a year.

StreamingHand gets bandwidth from Cogent. StreamingHand resells it to buyers with fair and transparent market-pricing. Traffic routing is carefully and conservatively controlled across multiple providers to avoid degrading quality, so for example, the Cogent circuit will never be loaded more than 40%. If Cogent can continue, the StreamingHand market will pass on the savings. If they go out of business, or raise prices, StreamingHand will simply re-adjust routing without affecting customer connectivity or quality. Rather than hiding behind a deceptive flat-rate scheme where buyers doesn't really know what they are getting, the market provides all buyers (for any level of traffic) the bandwidth they want at the best possible fair price.

What is a Tier 1 ISP? Background on ISP interconnection: peering and transit

  1. Transit service: the seller ISP carries your traffic to anywhere on the Internet.
  2. Peering: the ISP carries your traffic only if it is destined to its own transit customers.
A Tier 1 is a provider that can reach the whole Internet without buying transit from any other provider. If you are a Tier 1 ISP, every address on the Internet is either a transit customer of your network, or of one of your peers. If you are not a Tier 1, then you must buy transit from at least one Tier 1 in order to reach the whole Internet.