Moore's Law
Moore's Law: Computing power roughly doubles every two years at constant cost. This exponential improvement has driven 60 years of technological progress. Understanding the pattern β and recognizing that physical limits are now slowing it β is essential for technology strategy.
What Is Moore's Law?β
In 1965, Gordon Moore (then at Fairchild Semiconductor, later co-founder of Intel) observed that the number of transistors that could be placed on an integrated circuit had doubled approximately every year since 1959, and he predicted this would continue for at least 10 years. He later revised the doubling period to roughly two years.
The law is empirical β it's an observation about historical rates of change, not a physical law that governs chip design. But it became self-fulfilling: semiconductor companies used it as a planning target, organizing R&D roadmaps around the doubling goal. For approximately 60 years (1960sβ2010s), the industry tracked Moore's Law with remarkable fidelity.
The practical consequence: a smartphone contains more computing power than the most powerful supercomputers of the 1970s. The cost of a unit of computing has fallen by many orders of magnitude. This cost reduction enabled: personal computers, the internet, smartphones, cloud computing, genomics, machine learning, and most other technology revolutions of the past 50 years.
The broader "Moore's Law" pattern: The doubling pattern applies beyond transistors to storage density (hard drives, flash memory), network bandwidth, and solar cell efficiency. Each of these Moore's Law analogues has enabled its own wave of technological disruption.
Three Real-World Examplesβ
Cloud Computing Economicsβ
AWS and other cloud providers have been able to continuously lower prices for computing while maintaining margins because the underlying hardware costs (governed by Moore's Law) fell faster than prices. A computation that cost $1 on AWS in 2010 costs fractions of a cent today. This falling cost curve enabled business models (mobile apps, streaming video, machine learning as a service) that were economically impossible at earlier cost levels.
Technology strategy that assumes current computing cost structures will persist systematically underestimates the opportunity space. What is currently too expensive to compute economically often becomes viable within a product cycle.
The Sequencing of the Human Genomeβ
The Human Genome Project sequenced the first human genome from 1990β2003 at a cost of approximately $3 billion. By 2023, a human genome could be sequenced for under $200. This cost reduction β faster than Moore's Law, driven by specific innovations in sequencing technology β enabled an entirely new industry: consumer genomics, liquid biopsy cancer diagnostics, and personalized medicine.
This is Moore's Law applied to DNA sequencing (sometimes called Carlson's Law or Wright's Law in this context). The lesson: cost curves that decline exponentially create discontinuities in what is economically feasible at each order-of-magnitude threshold.
The Limits and the Post-Moore Eraβ
Physical limits to Moore's Law have become binding. Transistors are now measured in nanometers β approaching the scale of individual atoms. Quantum effects create noise; heat dissipation becomes unmanageable. The doubling rate for raw single-core performance has slowed significantly since roughly 2006, though progress continues through parallelism (multiple cores), specialized processors (GPUs, TPUs, AI accelerators), and architectural innovation.
The post-Moore era doesn't mean the end of computing progress β it means progress comes from different places. GPUs enabled deep learning not because of Moore's Law progress but because of architecture suited to parallel computation. Custom AI chips (Google's TPU, Apple's Neural Engine) provide improvements specific to AI workloads. Understanding that Moore's Law is bending changes technology strategy: raw compute scaling can no longer be assumed; efficiency and specialization matter more.
When to Use Itβ
β Apply Moore's Law thinking when:
- Building technology roadmaps that depend on computing cost or capability
- Evaluating when a currently expensive technology will become economically viable
- Analyzing the business model of companies whose profitability depends on hardware cost curves
- Planning infrastructure investment with 5+ year time horizons
| Pairs well with | Why |
|---|---|
| Power Laws | Exponential growth is a specific form of power law over time |
| Lindy Effect | Moore's Law beneficiaries accumulate advantage; Lindy explains incumbent persistence |
| Scenario Planning | Technology roadmaps under Moore's Law uncertainty benefit from scenario analysis |
Three Real-World Examplesβ
The Smartphone in Your Pocketβ
The iPhone 15 Pro (2023) contains an A17 processor with approximately 19 billion transistors on a 3nm process. The Intel 4004 (1971) had 2,300 transistors on a 10,000nm process. That's roughly an 8-million-fold increase in transistor count over 52 years, largely consistent with Moore's Law doubling every 18β24 months. This improvement is why a $800 phone has more computing power than a 1990s supercomputer that cost millions of dollars and filled a room.
The Collapse of DNA Sequencing Costsβ
The Human Genome Project (completed 2003) cost approximately $2.7 billion and took 13 years. By 2023, whole genome sequencing cost under $200 and could be completed in hours. This is faster-than-Moore's-Law improvement β sequencing cost has halved roughly every 7 months since 2008. The implication: personalised genomic medicine, previously a luxury of research institutions, is becoming routine clinical care. This cost trajectory was predictable by extrapolating the technology curve.
Solar Panel Cost Declineβ
Solar photovoltaic module prices have declined approximately 90% per decade since the 1970s β a pattern now called "Swanson's Law," the solar equivalent of Moore's Law. In 2010, utility-scale solar cost around $0.40/kWh; by 2023, costs had fallen below $0.03/kWh in the sunniest regions. This predictable cost curve enabled rational investment in solar infrastructure decades before it became the cheapest electricity source, because investors could extrapolate the trajectory with reasonable confidence.
When to Use Itβ
β Apply Moore's Law thinking when:
- Forecasting technology costs over 5β20 year horizons (compute, storage, bandwidth, genomics, solar)
- Evaluating whether a currently-uneconomic technology will become viable at scale
- Building business models that depend on future technology cost assumptions
β Be cautious:
- Moore's Law is slowing in traditional silicon transistor scaling β physical limits are being approached
- Not all technologies follow Moore's Law trajectories; batteries, for instance, have improved more slowly
- Extrapolating technology curves requires understanding the underlying physical and economic drivers
| Pairs well with | Why |
|---|---|
| Second-Order Effects | Technology cost declines produce second-order effects across dependent industries |
| Scale Effects | Moore's Law is partly driven by scale effects in chip manufacturing |
| Lindy Effect | For technologies following Moore's Law, Lindy applies to the trend, not the specific technology |
Common Misuses and Limitationsβ
Assuming Moore's Law is a physical law. It is not β it's an empirical observation and a self-fulfilling prophecy. The semiconductor industry planned R&D investment and fab construction around the roadmap, which made the prediction come true. The law persists partly because the industry organised itself to fulfil it.
Applying it to all technologies. Moore's Law applies to transistor density. It has analogues in storage (Kryder's Law) and bandwidth (Nielsen's Law) but not to all technologies. Battery energy density improves at roughly 5β8% per year β far slower than the 41%/year of transistor density. Transportation speed has barely changed in 50 years.
Ignoring the slowdown. Transistor scaling below 5nm is running into fundamental physical limits (quantum tunnelling, heat dissipation). The industry has shifted strategies: from transistor count to architectural improvements (parallelism, specialised chips, packaging innovation). The rate of improvement is slowing even if absolute progress continues.
Related Modelsβ
| Model | Relationship |
|---|---|
| Scale Effects | Manufacturing scale effects partially drive Moore's Law cost declines |
| Second-Order Effects | Technology cost declines create cascading second-order effects across industries |
| Power Laws | Exponential technology improvement creates power law gaps between technology generations |
Frequently Asked Questionsβ
Is Moore's Law dead?
Traditional transistor density scaling is slowing β physical limits at sub-5nm nodes are making each new process node harder and more expensive. But "computing performance per dollar" continues to improve through architectural innovation: multi-core processors, specialised AI chips (GPUs, TPUs), chiplet designs, and 3D stacking. The practical consequence of Moore's Law β that compute gets faster and cheaper β continues even as the mechanism shifts. The rate of improvement is slower than in peak decades, but the trajectory persists.
How should businesses use Moore's Law for strategic planning?
Model future technology costs by extrapolating current trajectories: if a capability costs $X today and has been declining at Y% per year, estimate future costs and ask "at what price point does this become viable/disruptive for our industry?" Companies that anticipated cheap genomic sequencing built diagnostics businesses. Companies that anticipated cheap solar built energy businesses. The insight is to act before the price point is reached, not after β competitive advantage requires anticipation, not reaction.
What is the relationship between Moore's Law and AI progress?
AI progress depends partly on compute (training larger models), which benefits from Moore's Law, and partly on algorithmic efficiency improvements, which have historically been faster than Moore's Law (AlexNet's 2012 ImageNet accuracy required roughly 100Γ less compute to achieve in 2020 than in 2012). The current AI boom has been enabled by the intersection of Moore's Law (cheap compute), large datasets, and algorithmic improvements that compound independently. Even if hardware improvement slows, algorithmic gains may sustain AI progress trajectories.
Further Readingβ
- Moore, G. (1965). "Cramming More Components onto Integrated Circuits." Electronics Magazine β the original paper
- Mack, C. (2011). "Fifty Years of Moore's Law." IEEE Transactions on Semiconductor Manufacturing
- Waldrop, M.M. (2016). "The Chips Are Down for Moore's Law." Nature
Apply with AIβ
π Apply Moore's Law to your technology strategy in MindMax β
This page is part of the MindMax Mental Models Knowledge Base.