What does Moore's Law really teach us?

In 1965, Gordon Moore, then director of research and development at Fairchild Semiconductor, wrote a short article about the future of electronics. He observed that the number of components that could be placed economically on an integrated circuit had been doubling roughly every year and projected that the trend could continue for another decade.

In 1975, he revised the pace to approximately a doubling every two years. The observation later became known as Moore's Law.

Despite its name, it is not a law of nature. It does not operate with the inevitability of gravity. It was an extrapolation from evidence available at the time. Yet the trajectory remained remarkably useful for decades.

Moore's Law mattered not only because it anticipated the future, but because an entire industry organized itself to build that future.

The force of repeated doubling

Linear improvement is intuitive: add the same amount each year. Repeated doubling appears modest at first and then creates an enormous difference.

Starting from one, ten doublings produce 1,024. Twenty produce more than one million. When capability compounds this way, a small improvement today can eventually change the structure of an industry.

As more transistors fitted onto a chip, computing capability increased and the cost of each unit of computation fell. Capabilities once confined to institutional laboratories reached personal computers and then phones held in the hand. Modern AI rests partly on this accumulation of semiconductor progress.

But twice as many transistors do not guarantee twice the useful performance. Software, memory, energy, heat, networks, data and the needs of users shape the final result. “Computers become twice as fast every two years” is therefore an oversimplification of Moore's original observation.

How did an observation become an industry target?

Chip designers could not sustain this trajectory alone. Materials science, manufacturing equipment, lithography, precision measurement, software, investment and a global supply chain all had to advance together.

Companies planned factories around the expected arrival of the next generation. Suppliers developed the tools those factories would require. Researchers worked on the limits that current methods could not cross. An observation gradually became a shared expectation and then a coordinating roadmap.

Technological progress was not automatic. It was the product of decades of research, enormous capital, creative engineering and coordination across institutions.

A demanding direction does not guarantee the future. It aligns decisions so that people can build, step by step, what once appeared impossible.

Is Moore's Law over?

As transistors become smaller, physical and economic obstacles grow. Leakage, heat, manufacturing complexity and the rising cost of advanced fabrication make the old pace increasingly difficult to sustain.

This is why arguments about the “end” of Moore's Law continue. A simple yes or no misses the change underway. Traditional transistor scaling has slowed, while progress continues through new architectures, advanced packaging, specialised processors, memory, software and energy efficiency.

The exact numerical cadence now matters less than the habit of thought behind it. When one path reaches a limit, the problem can be reorganised at another level.

AI progress is not simply Moore's Law

AI capabilities have recently advanced at striking speed. Computing power is an important cause, but Moore's Law alone does not explain the change.

Model architectures, training methods, data quality, specialised chips, data centres, energy and software improvements combine to create current results. Multiplying compute does not automatically multiply intelligence, reliability or decision quality.

The distinction matters. Technical capability can improve rapidly. The rate at which an organisation applies it to the right problem, makes it accountable and embeds it in reliable work is different.

What does this mean for mining?

Mines are designed to operate for decades. A project planned today may still be producing in 2040, 2050 and beyond. During that life, computing, sensing, automation and AI may pass through many generations.

A long-life mine should therefore not be locked rigidly to today's technology. It needs foundations that preserve high-quality data, connect systems, allow safe experimentation, measure results and scale successful applications.

A more powerful computer does not interpret an orebody by itself. A better sensor does not change a mine plan. AI creates value only when it is connected to sound data, a clear purpose, professional judgement and real decision rights.

Even when equipment capability improves, mining productivity does not necessarily rise at the same pace. Roads, plans, maintenance, ore variability, work practices and coordination constrain the whole system. This is the gap between technological potential and organisational capability.

As technology improves faster, the limiting factor may become not the machine, but the organisation's ability to turn new possibility into dependable work.

The lesson for Mongolia

Mongolia does not need to lead every field of semiconductor manufacturing. As computing becomes more accessible, the country gains a greater opportunity to solve problems it understands deeply at a global standard.

Mining, extreme climate, remote infrastructure, energy, water and logistics are domains of lived experience. Solutions to these challenges can become more than one-off improvements at one site. They can become data products, software, engineering methods, intellectual property and exportable companies.

Buying technology is not the same as owning capability. Mongolian engineers need to define problems, govern their data, test solutions, prove outcomes and improve the next version themselves.

The deepest lesson from Moore's Law is not speed but accumulation. National capability is more likely to come from improvement repeated every year, supported by patient investment, than from one dramatic leap.

Questions for an organisation

Are we merely buying technology, or using it to build a capability of our own?

Will the data we create today remain useful for decisions five or ten years from now?

Do we have a clear path from experiment to dependable everyday use?

Which constraints in people, process and governance keep us from learning at the pace of technological change?

The answers may say more about an organisation's future than the specification of its newest equipment.

The idea behind the speed

Moore's Law does not promise that everything will double forever. Every technology meets physical, economic and social constraints.

It does show that when people commit to a measurable direction and align knowledge, capital, manufacturing and supply chains around it, they can transform the boundary of possibility within a few decades.

For us, the most important question is not how many transistors the next chip will contain.

As technological possibility compounds, is our capacity to learn, coordinate and turn it into real value compounding too?

About the sources and this interpretation

The history of the original observation and its 1975 revision draws on Gordon Moore's 1965 article “Cramming More Components onto Integrated Circuits,” its IEEE reprint and Intel's historical account. The discussion of present limits and paths beyond traditional scaling was compared with material from IEEE and imec. The connections to mining, organisational learning and Mongolia's opportunity are my interpretation.