A French bank hopes that bundling spare capacity in hundreds of its computers will give it an edge in the financial markets.
Societe Generale is launching the latest application of "grid computing", which pools unused computer functions and memory across a network to create what amounts to a supercomputer.
Grid computing has already been used to crunch huge volumes of complex data in scientific research, notably in biochemistry and astronomy.
Now - in a crucial test of the value of the technique - it is being tried out in the commercial world.
Societe Generale is launching a pilot grid computing project next month at its commodity trading division, which deals in everything from coffee to platinum.
The bank has joined forces with grid firm Platform Computing to speed up the pricing of financial instruments such as futures and options, a task that requires notoriously tricky computation.
By spreading the job across up to 500 computers in its offices around the world, Societe Generale hopes to make sure its staff get the data in as close to real time as possible.
The other area where the financial sector is trying out grid computing is in the back-office sector, where millions of transactions have to be processed and checked every day.
Royal Bank of Canada has used the technique in its insurance division, reducing a 2.5-hour routine job to just 10 minutes.
But grid computing as a genuine commercial tool is still in its infancy, primarily because the logistics of redeploying computer capacity are too complex for most companies to bother with.
Nonetheless, proponents of the technique argue that it could eventually become a substitute for conventional computing.
"The idea has been around for 20 years," Dominique Comte, Platform's Managing Director for Southern Europe, told BBC News Online.
"But now we have the internet, and networks capable of interconnecting as they can today, it has become practical."
The market for grid services could reach $5bn by 2008, up from $250m this year, consultancy Insight Research has forecast.