Moore's Law and its limits
The Prophecy That Built the Modern World — and Why It's Running Out of Room
The most consequential prediction in the history of technology was made by a man staring at a graph he'd sketched on a piece of paper in 1965.
The Idea
Gordon Moore, co-founder of Intel, noticed something strange: the number of transistors engineers could fit on a silicon chip had been doubling roughly every two years, and costs had been halving in step. He extrapolated this into a prediction — and then, by working at the companies that shaped the industry, helped make that prediction come true. Moore's Law isn't a law of physics. It's closer to a self-fulfilling industrial roadmap, a collective commitment by chipmakers, equipment manufacturers, and software developers to keep pace with each other. For fifty years, it held. The result was exponential improvement so consistent that we stopped noticing it — smartphones in every pocket, genomic sequencing for the price of a doctor's visit, weather forecasting that would have seemed like sorcery to a 1970s meteorologist. But the physics is now fighting back. Transistors are approaching the size of individual atoms — the latest chips pack features measured in just a few nanometres, a nanometre being roughly the width of ten hydrogen atoms. At these scales, electrons stop behaving predictably. Quantum tunnelling — where particles leak through barriers they classically shouldn't be able to cross — causes chips to generate heat and make errors in ways that can't simply be engineered away. The doublings are slowing, the costs of each new generation are rising sharply, and the industry is quietly rewriting the story it tells about progress.
In the World
In 2022, TSMC — the Taiwanese company that manufactures chips for Apple, Nvidia, and most of the world's cutting-edge devices — began shipping its 3-nanometre process. The engineering achievement was staggering: roughly 292 million transistors packed into a single square millimetre. But the factory required to produce it cost around 20 billion in construction alone, and only a handful of companies on earth could afford to design chips complex enough to justify using it. This points to the real shape of Moore's Law's decline. It isn't a sudden wall but a gradual narrowing. Progress continues, but it concentrates. The doubling still happens — just more slowly, more expensively, and for a shrinking club of players. The democratising force that once let a scrappy startup outpace an established giant by waiting a generation for cheaper hardware has weakened considerably. The industry's response has been creative: rather than making individual transistors smaller, chipmakers are now stacking chips in three dimensions, specialising processors for specific tasks (the GPU boom powering AI is a direct example), and experimenting with entirely new materials beyond silicon — gallium nitride, silicon carbide, even carbon nanotubes. The end of Moore's Law as originally conceived is also, in a strange way, the beginning of a much more interesting and fragmented era of hardware innovation.
Why It Matters
The slowdown in Moore's Law isn't just a story for hardware engineers — it quietly reshapes assumptions embedded in almost every field that depends on computation. For decades, software developers could write inefficient code and trust that next year's hardware would compensate. That free lunch is ending. Efficiency in software is starting to matter again in ways it hasn't since the 1980s. The cost of running large AI models, for instance, is becoming a genuine economic and environmental constraint — not a footnote. At a broader level, Moore's Law was one of the few mechanisms in modern life that reliably delivered more for less, year after year. Its slowing invites a harder question: where does the next compounding curve of improvement come from? Quantum computing is a candidate, but remains largely pre-commercial. Biological computing is another. The honest answer is that nobody knows — and the gap between the end of one exponential and the start of the next one is exactly where the interesting bets are being placed right now.
A Question to Ponder
If cheap, reliable computation had been quietly subsidising progress in dozens of fields for fifty years without anyone fully accounting for it, which of those fields is most exposed now that the subsidy is shrinking?
Get a new one of these every morning.
Start learning with Thinkable