So, what is the good of a market forecast that you can drive a truck through? To answer this, it helps to understand what market research is and what is it is not. Good market research does two things. First, good market research fully encapsulates current expectations. Second, it adapts rapidly as expectations change. Market research is not intended to be a crystal ball.

Truly outstanding market research not only encapsulates current expectations and changes rapidly as expectations do, but it also provides a confidence interval.

Now, how does all of this pertain to Infiniband? The numbers that I have calculated-a range between $300 million and $1 billion-represent a 95 percent confidence interval for the Infiniband silicon TAM. The interval is wide because there is so much uncertainty about Infiniband deployment, with opinions basically diverging into two deployment scenarios.

I call the first group the "infrared camp." These folks believe that by 2006, nearly 100 percent of servers will ship Infiniband-ready, an idea that's reminiscent of infrared ports on PCs. Nearly all PCs have them, but nobody cares except for infrared-silicon vendors. Under the infrared assumption of ubiquity, the Infiniband silicon market should exceed $1 billion.

The other group might be called the "Fibre Channel camp." Using Fibre Channel as an analog for Infiniband, this camp assumes that Infiniband is deployed only where it will be enabled-specifically, in only those server clusters that use Infiniband and in select storage-area networks.

As uncertainty diminishes over time, this extremely wide confidence interval will narrow. It is simply too early to say in which direction-infrared or Fibre Channel-and thus to predict with any confidence whether the Infiniband silicon market will be worth $300 million or $1 billion. The truth probably lies somewhere in between.

Jeremey donovan (jeremey.donovan@gartner.com) is a principal analyst at gartner dataquest.