Milking Moore's Law

Computerworld - Moore's Law has been going strong for decades. And for all that time, IT has kept sopping up the resulting compute power, on clients and servers alike. Not for nothing is it said, "Intel giveth and Microsoft taketh away."
So will this situation continue indefinitely? In a word, yes. Demand for processing power won't be saturated for many years, though online transaction processing may actually be the one major exception to this trend. Growth in classic transactions is tied to real-world factors such as revenue, units sold or head count. Only rarely do these grow at Moore's Law speeds. Nontraditional types of transaction processing -- such as Web site logging and, soon, RFID -- give periodic boosts to overall transaction volumes, but their growth soon levels off.
But the analysis of that transaction data is surely on a much steeper curve. Today's corporations typically have BI systems that provide high volumes of canned, repetitive reporting but relatively low volumes of personalized information delivery. When every manager in the organization has a personalized dashboard and uses it for frequent drill-down analysis, analytical processing requirements will explode.
An even greater explosion in analytic processing could come in planning. Ideally, one would like to test various assumptions about the future more or less independently of one another. This can create a huge number of planning scenarios, limited only by the available computing power, which in turn may be limited only by the questions of what the economic value of superior planning really is.
Another area of near-infinite potential growth in processing is predictive analytics. There's no real limit -- except imagination -- to the amount of data mining that can be attempted, at least for enterprises with large consumer markets. And when mining uncovers a useful nugget, it may be turned into an in-line model used in real-time customer interactions, potentially another huge use of computing power.
The most significant growth in server-side processing requirements, however, may come in a different area altogether: document information retrieval. Enterprise effectiveness would be greatly improved if computers could extract information directly from e-mail archives or even from a corpus of well-edited articles.
In principle, one could teach a computer to apply all the document understanding techniques humans use, only faster. But in practice, that will never happen, because it's too complex a programming challenge. Instead, the eventual solutions will be similar in nature to championship-level chess algorithms -- they will apply a small subset of the heuristics humans use, which will narrow the problem enough for