Why did computer chips double in power every two years, and can it last?

Mostly because chipmakers kept shrinking transistors, and for decades a smaller transistor was also faster and no hotter. Moore's law is the observation that the number of transistors on a chip doubles about every two years, at little extra cost. It is not a law of physics. In 1965 Gordon Moore, head of research at Fairchild Semiconductor, was asked by Electronics magazine to predict the next ten years. Components per chip had been doubling every year, he noted, and he bet it would go on. In 2015 he admitted it was 'just a wild extrapolation.' In 1975 he revised the pace to every two years.

Two things kept the curve going. The first was physics that played along. In 1974, IBM's Robert Dennard showed that as transistors shrank, their power use per area stayed constant, so chips could pack in more switches without running hotter. The second was people. The industry took Moore's prediction as a target for planning and research, and rivals raced to meet it, until it was seen as a self-fulfilling prophecy. Today a chip the size of a fingernail holds billions of transistors, with millions of times the capacity of early-1970s chips.

Can it last? Part of it already ended. Around 2005 to 2007, Dennard scaling broke down: leakage current stopped shrinking along with transistors, so power density began to climb. Clock speeds stalled at 4 to 6 gigahertz, because going higher would overheat the chip. And since around 2010, chipmakers report that progress has slowed slightly below Moore's pace.

Their answer was to stop trying to make one core faster and to put several cores on the same chip, splitting the work between them. That is why 'how many cores' became as important a question as clock speed.