One of the great things about technology is that it’s surprisingly easy to underestimate how strong the demand for it can become. Remember IBM CEO Thomas Watson’s purported 1943 remark, “There’s a world market for maybe 5 computers,” and Bill Gates’ supposed 1980 comment, “640 kbytes of RAM ought to be enough for anyone”?

Interestingly, there’s no evidence that either is an actual quotation -- and Gates in particular has emphatically denied the words attributed to him. However, they linger in the public consciousness as a way of reminding ourselves of a greater truth: The appetite for technology can surprise even seasoned veterans.

That certainly happened to me recently. Awhile back, my colleagues and I decided to look at how demand for Internet capacity might change over the next five years, and see whether the existing and planned infrastructure is adequate. To do so, we modeled user consumption of bandwidth over time, validated it against the best available data and then projected that demand forward. It’s essentially a Moore’s Law model of Internet demand, in that it looks at the rate of increase in a commodity (processing power in the case of Moore’s Law, and bandwidth use here). Then we compared that demand to existing and planned infrastructure capacity (see our report here).

This differs from the typical approaches to measuring Internet growth. One common tactic is to measure, as closely as possible, the traffic on the Internet core and project most recent growth trends. The catch here is that, by definition, measuring traffic on the core looks only at packets that have made it onto the Internet. So it’s not an effective measure of user demand, because packets could be gated at the edges.

The other approach is to look at Internet applications and see how quickly they’re being deployed, and project those use patterns into the future. But by definition, this tracks only applications that we know about today — and therefore would miss a phenomenon such as YouTube, which was unknown before 2005, yet generated an estimated 27 petabytes per month by mid-2007.

In contrast, looking at use of local bandwidth remedies both issues. It provides a good way to measure how users would consume bandwidth if it were available without requiring a crystal ball to envision not-yet-invented applications.