Demand is just a river in (oh, wait)…

Arnold Kling takes issue with my comment that Worldcom’s collapse means Moore’s Law has come to telecom. (On the other hand, Brad DeLong agrees with me, though he doesn’t realize it.)

Arnold says Worldcom’s problem was over-estimation of demand, not over-paying for supply. They built a network for 1000x annual traffic growth, but actual growth was 27x. What Arnold misses is that it would be the same network. The big capital expenditure was putting in the fiber and lighting it up. Worldcom had to do that even under the lower growth expectaitons. Swapping out faster gear is a relatively minor minor cost component, because (and here’s the punch line) networking gear follows Moore’s Law in price/performance improvements.

When I say Moore’s Law comes to telecom, here’s what I mean. The IBM laptop I bought 3 years ago cost $3,000. (Actually, I got it for free, but that’s neither here nor there.) Today, that model is available for $700 on eBay. That’s about the same depreciation rate that WorldCom’s network experienced. Here’s the difference. The PC industry can sell me a new laptop today for $2,000 or more, because it has more horsepower to run today’s demanding Microsoft software. Worldcom was stuck making only $700. A bit is a bit, after all. Brad’s point about capacity increases is essentially the same. Moore’s Law is why capacity can increase so dramatically.

The WorldCom demand hype tale is a good one, first thoroughly exposed by the ever-perceptive Andrew Odlyzko. It’s a little too neat, though. It was John Sidgmore of UUNet, not Bernie Ebbers, who made the claim most voiciferously. And he got it from Mike O’Dell, UUNet’s chief scientist. Mike was the guy actually doing network architecture planning. He knew what real demand curves were like, and what network investments cost. He made the 1000x growth statement back in 1995-1996, when even Odlyzko admits it was actually true. Sidgmore spent several more years spouting that number in speeches, and may have convinced Wall Street that his sow was a silk purse. But O’Dell was the one deciding what to put into the network.

Believe you me, Mike isn’t a hype artist or bulshitter. When I interviewed him for Release 1.0 in 1998, he tore me a new one for not focusing more on traffic engineering. That means using smart routers to manage network traffic efficiently, in contrast to the "throw bandwidth at the problem" mentality that was common at the time.

Worldcom made more than its share of mistakes. (We’re not even talking about the accounting fraud.) But the critical test they — and most Net other infrastructure providers — failed was their inability to cope with relentlessly declining costs.