Moore’s law is the observation that the number of transistors per integrated circuit doubles roughly every two years. It’s not a law of physics but a self-fulfilling industry roadmap: Gordon Moore made the original observation in 1965 (then predicting yearly doubling), revised it in 1975 to every two years, and the semiconductor industry has organized its R&D budgets around hitting it ever since. The popular “18 months” figure is a separate claim about performance attributed to David House of Intel — Moore’s own statement was about transistor count and two years.
The practical consequence is that every couple of years, the same chip area gets twice the logic, or the same logic shrinks to half the area. That’s why a modern smartphone has more compute than a 1990s supercomputer.
It’s been called dead many times — most recently because shrinking transistor gates below ~5nm runs into quantum tunneling and atomic-scale variation — and yet performance per chip keeps climbing, increasingly through 3D stacking, chiplets, and architectural tricks rather than pure feature shrinks.
Moore’s law is what makes FPGAs with millions of LUTs commercially viable, and it’s the implicit assumption behind every “we’ll just throw more compute at it” engineering decision.