Garry Kasparov, perhaps the most brilliant chess-playing mind the world has ever produced, looked shaken. An IBM computer, Deep Blue, had just beaten him over six games. He told reporters: "I'm a human being. When I see something that is well beyond my understanding, I'm afraid."

But that was 1997, and this is now. At its lab in Ruschlikon, a quiet suburb of Zurich, IBM has just unveiled its latest supercomputer. It makes Deep Blue look like a Ford Anglia parked beside a Lamborghini.

Now that it can dismiss the world chess champion, the scientists have turned the attention of their technology towards other tasks. With the sort of computing power available to them - 800 million calculations per second from 128 Power3 processors in 64 nodes and 64 gigabytes of RAM - these are many.

Among the more dramatic uses for "deep computing" is weather prediction. Meteorologists already use computers to process certain amounts of data, but on nothing like the scale possible with the fast processors. IBM has developed a system that sucks vast amounts of data into the computer, processes the information by comparing it to previous weather, spots millions of patterns, then predicts what conditions will be like.

This not only gives a more accurate weather forecast than currently available, it can also do so for a far smaller area. An early prototype of the system was pressed into action when a storm threatened to wreck the closing ceremony at the Atlanta Olympic Games in 1996. The organisers wanted to postpone the event, but were persuaded by IBM that the storm would by-pass the stadium by several kilometres. The computer got it right.

In Zurich this week, weather computing specialist Dr Zaphris Christidis said being able to predict the weather with such accuracy would prove a major boon to many businesses.

"Demand for electricity is closely linked with weather conditions such as temperature, so precise weather forecasting can be a powerful tool to help utilities plan production," he said. "For example, the timing of the peak high temperature during a summer day has a large effect on patterns of power demand."

Another way to use supercomputing to make business more efficient lies in scheduling. This is not a problem for all companies, but for airlines, for example, it represents a permanent headache. Even comparatively small carriers, like Aer Lingus or Ryanair, have to draw schedules that factor in maintenance, routing, flights schedules, aircraft type, crew locations and preferences, labour restrictions and other regulations.

Instead of taking months to prepare schedules - with consequent havoc when something goes awry - deep computing should allow airlines and companies with similar difficulties reconfigure all the variables in minutes.

"The systems also make it possible to model `what-if' scenarios," said Dr Avraham Harpaz, "to examine, for example, the cost impact of a particular change in the rules governing working hours."

El Al has already tried the system, and reduced its crew-related costs by around 6 per cent. But where supercomputing meets serious money is in the exploration sector. Looking for oil is now an extremely expensive business, and the shortage of sites on land is shifting the focus of exploration to remote - and therefore even more costly - deep-water locations.

Each well costs up to €50 million to drill; nine out of 10 wells are dry.

But now, according to Dr Ulisses Mello, IBM's deep computing team have created a simulation of the oil-rich Gulf of Mexico.

"As one part of the model generates a history of temperatures and pressures, a second part takes these results and simulates the formation of oil," he said. "The resulting 3D model of simulated oil maturity is in reasonably good agreement with what is known of oil distribution in the Gulf - perhaps accurate enough to allow an oil company to assume a most optimistic and most pessimistic scenario in deciding whether to drill in a particular spot."

For retailers too, deep computing could represent a boon. Already, an online bookshop might massage enough data about its customers' preferences to discover that people who like books by Elmore Leonard and Ed McBain will also probably enjoy the work of Carl Hiaasen. But with deep computing, they can discover buying patterns that no one can see now, allowing far more targeted, and lucrative, selling.