Skip to content
Neural NeerajNeural Neeraj
Blog
All posts
Your Phone Has More Transistors Than Humans on Earth

Your Phone Has More Transistors Than Humans on Earth

5 min read
HardwareAISemiconductors

13 Sextillion Transistors Have Been Manufactured -- 1,700x More Than Grains of Sand on Earth

That number broke my intuition. We've manufactured more computational switches than there are grains of sand on the entire planet, by a factor of 1,700. And we keep accelerating.

The Crossover Point

The Apple A18 Pro packs 20 billion transistors. Earth's population is approximately 8.28 billion humans. Your phone carries 2.4x more transistors than people alive.

The first chip to cross the human population threshold was Apple's A13 in 2019 -- 8.5 billion transistors versus 7.7 billion humans at the time. We blew past parity seven years ago and nobody threw a party.

Scale That Defeats Intuition

The numbers get absurd fast when you move beyond phones:

  • Apple M5 base chip (October 2025): 28 billion transistors
  • Apple M3 Ultra: 184 billion transistors -- roughly the number of stars in the Milky Way (~200 billion)

A single desktop chip approximating the star count of our galaxy. Each transistor is built at the 3-nanometer process node. A human red blood cell is about 7,000 nanometers across. A strand of DNA is 2.5 nanometers wide. These switches operate at molecular scale.

The Manufacturing Stack

Building 20 billion transistors on a chip requires ASML's EUV (Extreme Ultraviolet) lithography machines. Each machine costs ~$380 million, weighs 180 tons, and uses a 20-kilowatt laser to vaporize tin droplets 50,000 times per second. The resulting 13.5nm wavelength light passes through a mask containing the chip design and projects patterns onto photoresist-coated silicon wafers.

A modern chip requires roughly 80 lithographic steps and over 100 vertically stacked layers of interconnects, transistors, and metal routing. The precision involved: imagine standing in New York and using a laser pointer to etch consistent characters on a grain of rice in San Francisco. Now do it 20 billion times with zero variation.

Only one company on Earth builds these machines. ASML is a single point of failure for the entire semiconductor industry, and by extension, for AI.

Why This Is an AI Story

Large language models are a transistor story at their core. Training requires GPU clusters where each chip contains tens of billions of transistors. NVIDIA's H100 has 80 billion -- ten times Earth's population on a single die.

Inference is equally hungry. When you send a prompt to any LLM, your tokens are processed across billions of transistors switching states in coordinated patterns that produce coherent language. The irony is recursive: the AI systems pushing intelligence forward are entirely dependent on our ability to manufacture molecular-scale switches with near-zero defect rates.

Moore's Law: Slowing, Not Dead

Gordon Moore predicted in 1965 that transistor density would double roughly every two years. It held for five decades. We're hitting atomic limits now -- you can't make a transistor smaller than an atom -- but the industry has workarounds:

  • 3D stacking: Stack transistor layers vertically instead of shrinking horizontally (TSMC's 3DFabric)
  • Chiplets: Combine multiple smaller dies in one package instead of scaling a monolithic chip (AMD's EPYC approach)
  • Gate-all-around (GAA): Nanosheet transistors replacing FinFETs for better electrostatic control
  • Backside power delivery: Route power connections beneath the transistors, freeing the top layers for signal routing

These aren't incremental improvements. They're architectural pivots that buy another decade of scaling through engineering creativity rather than brute-force shrinkage.

The Perspective

Your phone contains more individual computational switches than there are humans who have ever used it. Each switch is smaller than a virus. They were manufactured by a machine that costs more than most countries' annual space budgets, built by a company that has a monopoly on a technology that makes the entire digital economy possible.

The hard question: at 13 sextillion transistors manufactured and accelerating, what's the energy ceiling? TSMC's fabs already consume more electricity than some small countries. The atoms are getting smaller, but the power bill isn't.