Moore's Law and the Slowdown.

In 1965, Gordon Moore — co-founder of Intel — published a four-page article in Electronics magazine observing that the number of transistors on a microchip was doubling every year. He revised the estimate in 1975 to doubling every two years, and that became the canonical version.

A few things people often get wrong about it:

• It was an observation, not a law of physics. Moore was extrapolating from data points. Nothing in nature requires transistor counts to double on schedule.
• It was about transistor density, not speed. The popular reading — that computers double in speed every two years — is a sloppy summary that conflates several distinct trends.
• It said nothing about cost or power consumption. Those were governed by a separate principle: Dennard scaling (Robert Dennard, IBM, 1974), which said that as transistors got smaller, their power density stayed constant — meaning you could pack more transistors and run them faster and keep heat in check, all at once.

For about thirty years, both held simultaneously. Computers really did get dramatically faster, cheaper and more power-efficient on a predictable cadence. Replacement cycles of two to three years could be justified on pure performance grounds: a new machine genuinely outclassed the old one.

ASML's re-released 2014 interview with Gordon Moore (1929–2023) on the prediction he made in 1965. Four minutes of the man himself reflecting on how a casual extrapolation became the guiding principle of the semiconductor industry — and why even he was surprised by its staying power.

Dennard scaling broke first, around 2005–2006. As transistors shrank below ~65 nm, leakage currents and quantum effects meant that smaller transistors no longer used proportionally less power. The free lunch was over: you could keep adding transistors, but you couldn't keep cranking up the clock speed without melting the chip.

That's why clock speeds plateaued around 3–4 GHz in the mid-2000s and have barely moved since. The industry's response was to add more cores (parallelism) rather than faster cores. That helps for some workloads (video encoding, scientific computing) but does almost nothing for the single-threaded code that dominates everyday office work and web browsing.

Moore's Law itself — pure transistor density — held longer, but has been visibly slowing since around 2015. The economic side of it has slowed even more: each new manufacturing node now costs multiples of the previous one. A leading-edge fab in 2026 costs 20+ billion euros to build. A 2 nm wafer costs roughly 4× a 7 nm wafer. The cost-per-transistor curve, which used to fall steadily, has flattened or even reversed at the leading edge.

Indicative single-threaded performance gains across CPU generations, normalised against 2005 = 100. The pre-2005 era doubled every two years; the post-2005 era is roughly flat by comparison.

If you bought a laptop in 2005, by 2010 it felt genuinely slow. The same workflow that ran smoothly five years earlier would crawl on the same hardware, and a new machine would feel transformatively faster.

If you bought a laptop in 2018, by 2026 it does not feel slow in the same way — at least not because of the CPU. A 2018 ThinkPad with 16 GB of RAM and an SSD will run Microsoft 365, Teams, Chrome with a dozen tabs, and Visual Studio Code roughly as well as a 2026 model of the same tier. The CPU is not the bottleneck.

What does age:

• Battery health — usually below 70 % of original capacity by year 4–5. Replaceable for €40–90.
• Storage — older laptops with mechanical HDDs feel obsolete, but an SSD upgrade transforms them.
• RAM — pre-2018 machines often shipped with 4–8 GB; modern web apps assume 16. Upgradable on most machines.
• OS support — Windows 11 dropped support for many pre-2018 CPUs. Linux distributions and ChromeOS Flex pick up where Microsoft drops off.

None of these are CPU problems. They are battery, storage, RAM, and licensing problems — and all of them are addressable for a fraction of the cost of a new device.

The slowdown of Moore's Law is the technical reason the 4 R's framework works in 2026. If hardware were still doubling in capability every two years, repair and reuse would be a losing battle: even a perfectly working three-year-old machine would feel hopelessly outdated.

But that's not the world we're in anymore. Hardware has plateaued. The bottleneck for most users is no longer compute power — it's component aging (battery), storage type (HDD vs SSD), and software bloat. All three are fixable. None of them require a new laptop.

The next lesson looks at why this matters environmentally: the carbon cost of a new laptop is overwhelmingly in its manufacture, not its use. Once you know that the new device isn't even meaningfully faster, the case for keeping the old one becomes both technical and ecological.