# Moore's Law
Gordon Moore sits at his desk in 1965... pen in hand... staring at a graph.
He's plotting transistor counts over time, and the line isn't just going up. It's *curving.* Exponential.
He's about to publish an observation that will become the most accurate prediction in tech history. Not because he discovered some law of physics. But because once he said it out loud... the entire semiconductor industry decided to make it true.
Think about that for a second.
Moore didn't find a natural law like gravity. He made a prophecy... and then thousands of engineers spent sixty years *willing it into existence.*
Imagine if an economist in 1965 predicted the stock market would double every two years... and then every investor on Earth coordinated to actually make it happen. That doesn't happen.
Except in semiconductors? It did.
Moore's working at Fairchild Semiconductor — this scrappy company in Mountain View that becomes the Cambrian explosion of Silicon Valley. Eight engineers had left Shockley Semiconductor in 1957 because their boss was... to put it kindly... impossible. They started Fairchild. And from Fairchild came Intel, AMD, and about fifty other companies that built the modern world.
Moore writes this article for *Electronics* magazine. Not *Nature.* Not *Science.* A trade publication most people threw away.
His prediction: the number of transistors you can fit on a microchip will double every year. Costs will halve.
He's not describing what *must* happen. He's describing what *could* happen... if everyone keeps pushing.
And they did.
For six decades.
We're talking about a prediction that held through personal computers... the internet... smartphones... AI. Name a technology that changed your life in the last fifty years — Moore's Law is somewhere in its bloodstream.
But let's get specific.
A transistor is a tiny switch. On or off. One or zero. More switches... more calculations.
In 1971, Intel releases the 4004 microprocessor. First commercial microprocessor ever. Twenty-three hundred transistors. Runs at seven hundred forty kilohertz.
Federico Faggin, the lead designer, hand-drew every single transistor path using red and blue pencils on enormous sheets of film... four hundred times larger than the final chip.
Picture this: You're an engineer in a lab in Santa Clara. Fluorescent lights buzzing. Coffee going cold. Drawing the future with colored pencils. Testing equipment humming in the background.
And when they power it on and it *works*... they know they're holding a new kind of fire.
Fast forward to the year 2000. Intel's Pentium 4 has forty-two *million* transistors. That's an eighteen-thousand-fold increase in less than thirty years.
But here's what makes your brain hurt: that growth isn't linear. It's exponential. Which means the second half happened way faster than the first. The jump from twenty-one million to forty-two million took less time than the jump from twenty-three hundred to forty-six hundred.
The curve gets steeper.
And here's where it gets wild.
Cost per transistor in 1971? About a dollar.
By 2020? Less than a *billionth* of a dollar. Not a millionth. A *billionth.*
You know what that means?
In 1975, a pocket calculator cost fifty bucks — real money back then, like two hundred eighty dollars today. It could add, subtract, multiply, divide. That's it.
Now you've got a phone in your pocket with more processing power than the Cray-2 supercomputer from 1985... and you use it to look at pictures of cats.
The computing power that would've cost millions in the seventies now costs pocket change.
Your phone has more processing power than all of NASA in 1969. We put humans on the moon with sixty-four kilobytes of memory. Your phone's background processes use more than that.
So here's the turn.
In 1975, Moore revises his prediction. Turns out doubling every year was optimistic. He adjusts it to every *two* years.
And this version becomes the industry gospel.
But notice what's happening. Moore's Law isn't like gravity. It's not something you discover and observe. It's something you observe... and then *chase.*
Intel, AMD, IBM — they start planning R&D cycles around hitting that doubling target. Self-fulfilling prophecy. Roadmaps get drawn. Budgets allocated. Careers rise and fall based on whether your team can shrink transistors by the expected amount.
Which raises this question: Is Moore's Law a description of reality... or a collective hallucination that we made real through obsessive effort?
The answer is yes. Both.
And that's more interesting than either option alone.
There's a parallel framework worth knowing.
In 1974, Robert Dennard at IBM figures out something crucial. As transistors shrink, their power density stays constant. You can make them smaller and faster without the chip turning into a tiny furnace.
This is Dennard Scaling. And for about thirty years, it's the secret sauce that makes Moore's Law possible. You're not just cramming more transistors onto a chip — you're doing it without melting the thing.
It's like discovering that as you pack more people into an elevator... each person weighs less. Doesn't make physical sense. But in semiconductors, for a while, it worked.
But then... mid two-thousands... Dennard Scaling breaks down.
Transistors are getting so small that quantum effects start messing with them. When your transistor gate is only a few atoms thick, electrons start doing this thing called quantum tunneling — they just *jump* across barriers they're not supposed to cross.
You're trying to build a wall out of fog.
Heat becomes a problem. Power consumption becomes a problem. The industry hits a wall.
So what do they do?
They pivot.
Two thousand five: Intel and AMD start releasing multi-core processors. Instead of one really fast brain, you get multiple brains working in parallel. Different strategy... but still honoring the spirit of Moore's Law. Keep doubling performance. Just find another way.
And this is where it gets philosophically interesting. The industry didn't say "well, we had a good run." They said "the law must continue"... and then they bent reality until it did.
Let's zoom out.
The Cray-2 supercomputer in 1985 was the fastest computer on Earth. Size of a refrigerator. Weighed fifty-five hundred pounds. Required liquid cooling with actual fluorinert cascading down its surfaces like a waterfall. Cost seventeen million dollars.
Seymour Cray, the designer, was this legendary recluse who did his best thinking while digging tunnels under his house. Seriously. He'd hit a design problem... go dig a tunnel... and the answer would come to him underground.
The Cray-2 could do one-point-nine gigaflops — nearly two billion calculations per second.
Today? Your laptop is more powerful.
Your *laptop.*
That shift happened because of Moore's Law.
But it's not just about speed. It's about what becomes *possible* when computation gets cheap enough.
DNA sequencing used to take years and cost billions. The Human Genome Project, finished in 2003, cost two-point-seven billion dollars and took thirteen years.
Now it takes hours and costs hundreds.
Why? Because the computers processing that genetic data got exponentially more powerful.
Same with astronomy. The James Webb Space Telescope generates fifty-seven gigabytes of data every day. The chips analyzing that data... finding galaxies that formed three hundred million years after the Big Bang... they're direct descendants of that 1965 prediction.
We're seeing the infant universe because transistors got small enough.
And AI. *Damn.*
Every generative model... every neural network... every time ChatGPT writes you a poem or DALL-E makes you an image — that's riding on Moore's Law.
GPT-4 was trained on something like twenty-five thousand NVIDIA A100 GPUs. Each one contains fifty-four billion transistors. The training run cost tens of millions in compute alone.
That's only economically feasible because Moore's Law made each calculation absurdly cheap. Without that exponential curve, AI would still be a curiosity in research labs... not something rewriting your emails.
Here's a connection nobody makes: Moore's Law is the closest thing we have to biological evolution happening in real time.
Natural selection takes millions of years to double complexity. Moore's Law does it every two years.
And just like evolution, it's not smooth — it's punctuated equilibrium. Long periods of incremental progress... then sudden leaps when someone figures out a new trick. FinFET transistors in 2011. Extreme ultraviolet lithography in 2018. Each one a Cambrian explosion in miniature.
But here's where it gets strange.
Around 2015, the fiftieth anniversary of Moore's original article, the industry throws a party. And at that party, everyone's whispering the same question: Is this thing finally dying?
Because the physics are getting brutal.
You can only make transistors so small before you hit atomic limits. A silicon atom is point-two nanometers wide. We're building transistors that are only ten atoms across.
At that scale, weird stuff happens. Quantum tunneling. Electron interference. You're trying to build a clock out of fog and probability.
Twenty twenty-three: Intel announces chips using twenty-angstrom technology. That's two nanometers. For context, a strand of human DNA is about two nanometers wide. They're building transistors the width of DNA.
IBM's doing similar work. TSMC is making three-nanometer chips for Apple's latest processors. Companies are using chiplet architectures — modular pieces that work together instead of one giant monolithic chip. The difference between building a skyscraper as one solid block... versus assembling it from prefab modules.
The law isn't dead. It's shape-shifting.
There's a debate worth sitting with.
Some engineers say we're done — physical limits, heat constraints, diminishing returns.
Others say we're just entering a new phase. Quantum computing. Neuromorphic chips that mimic how brains work. New materials like graphene or molybdenum disulfide.
The pessimists point to energy efficiency. Yes, we're still adding transistors... but the energy gains have plateaued since the mid two-thousands. Your phone is powerful, but it also needs charging every day. That's a Moore's Law problem we haven't solved.
Data centers now consume about one percent of global electricity. If AI keeps growing, that number could hit three or four percent by 2030.
We're approaching a world where the infrastructure to run our digital lives competes with the infrastructure to run our *actual* lives.
And there's an ethical angle nobody talks about enough.
Every time we upgrade, the old chips become e-waste. Fifty million tons of electronic waste generated every year. That's the weight of all commercial aircraft ever made... discarded annually.
Rare earth minerals get mined for new processors — often in places with minimal labor protections. Cobalt from Congo. Lithium from Chile.
Moore's Law gave us the future. But it also gave us a planet full of discarded electronics and an energy consumption problem that's only getting worse.
The circular economy people and the semiconductor industry are barely talking to each other. And that's a problem.
Picture a modern semiconductor fabrication plant. A "fab."
It's 2023, and you're standing in a cleanroom so sterile that a single speck of dust could ruin a million-dollar batch. The air in here is ten thousand times cleaner than a hospital operating room.
You're wearing a full-body suit. Face covered. Hands gloved. You can't bring in paper — it sheds particles.
The air smells like nothing. Filtered to perfection.
Machines hum quietly, etching circuits onto silicon wafers using extreme ultraviolet light with a wavelength of thirteen-point-five nanometers. That light is so energetic it would be absorbed by air... so the whole process happens in a vacuum.
You're watching two-nanometer chips being born. Each one contains billions of transistors. *Billions.*
And you realize: This is the most complex manufacturing process humanity has ever invented.
We can build these chips... but barely. We're at the edge of what's physically possible.
ASML — the only company that makes EUV lithography machines — ships maybe fifty units a year. Each one costs a hundred fifty million dollars and requires forty shipping containers to transport.
We're building temples to Moore's Law.
Gordon Moore, in that 1965 article, wrote: "The future of integrated electronics is the future of electronics itself."
He was right. But maybe not in the way he thought.
The future isn't just about making things smaller. It's about what we do with the power we've unlocked.
Moore died in 2023, at ninety-four... having watched his throwaway prediction become the organizing principle of an entire century.
So here's what you can take from this.
Next time you're frustrated that your phone is slow or your computer is old... remember: you're holding a device that would've been considered magic fifty years ago.
But also remember: the exponential curve that got us here might not keep going forever. Which means the next breakthroughs won't come from just making things smaller. They'll come from thinking differently about what computation even *is.*
And maybe that's the real lesson.
Moore's Law isn't about transistors. It's about what happens when you set an ambitious target... and then organize an entire industry around chasing it. The power of a shared obsession.
So ask yourself: What's *your* Moore's Law?
What's the exponential curve you could draw for yourself — not in technology, but in your work... your learning... your life? What would double every two years if you decided it would?
Because Moore's Law teaches us this: Sometimes the prophecy creates the reality. Sometimes saying it out loud is the first step to making it true.