Intel's business model rides on this law, using each new generation of processors to finance the research, development and fabs required for the next. A few years back Craig Barrett, then chief executive of Intel, jumped down my throat at a London press conference when I asked if the demand for processing power would keep pace.

His irritation was understandable: it is an old question, but also a sensitive one. Barrett harrumphed that people would always need more processing power.

This is true of server farms, and demanding niche applications. Yet by far the majority of users now take processing power for granted. Entry-level machines have more than enough for what most people use them for most of the time. Machines are bought on looks and price as much as on technology.

Hardware has had more than enough power for contemporary mainstream software for at least five years. Sales have grown because the global market has grown, but people and companies are not changing machines so quickly because they do not need to.

Barrett was right in that the demand for processing power will increase. Extreme gamers will exploit all that they can get; relatively few people are editing high-definition video, but their numbers will rise. Future interfaces exploiting multi-modal input will also push up demand, especially for energy-efficient mobile processors.

The question, which Barrett evaded, is whether this demand will keep step with Intel's business cycle. Or to put that another way: will demand for innovative technology be enough in itself, year after expensive year, to provide a return on the investment required to produce it? The assumption underlying Moore's Law is that it will.

Yet for the past few years, arguably, it has been competition rather than market demand that has prodded Intel into its biggest technology changes.

Maverick chipmaker Transmeta forced it to place a new priority on power consumption, though to be sure this would have become an issue in time. More recently AMD has been setting the pace, offering superior products that forced Intel to up its game.

Meanwhile processors are selling on price and efficiency as much as speed. Some increasingly important uses, such as set-top boxes, media servers and network-attached storage, need only modest computing power. AMD makes good processors and there is a lot of market for it to exploit. And, as it has shown more than once, it can outsmart Intel.

Oddly, whether Moore's law holds good for another 15 years could depend on AMD. Without competition, Intel could get lazy again.