54 Years of Data · May 2026

The Law That

Ate The World

2,300 → 184 Billion54 Years of DataProcess: 10µm → 2nmDoublings Every 2 Yrs

Peak Transistors

0B

Apple M3 Ultra, 2024

Process Shrinkage

0×

10µm (1971) → 2nm (2024)

Years Running

0 yrs

Intel 4004 to today

Transistor Growth

0M×

2,300 → 184 billion

🔵 Intel 4004 (1971) — 2,300 transistors — first commercial CPU🟢 Apple M3 Ultra — 184 billion transistors — 80M× more than 4004⚡ Nvidia B200 (2024) — 208 billion transistors — Blackwell AI beast📉 Process node — 10,000nm → 2nm — 5,000× smaller transistors🔮 Intel 2030 target — 1 trillion transistors — angstrom era💚 AMD EPYC Rome — 39.5B transistors — chiplet revolution (2019)🔴 Moore's Law slowing — 3-year doublings now — was every 2 years💡 Pentium (1993) — 3.1 million transistors — first desktop revolution🟡 Core 2 Duo (2006) — 291 million transistors — multi-core era🔵 Apple M1 (2020) — 16 billion transistors — ARM conquered Intel⚡ Nvidia H100 (2022) — 80 billion transistors — the AI gold-rush chip🟢 TSMC 2nm — volume production 2025 — pushing physics to the limit🔵 Cerebras WSE-3 — 4 trillion transistors — wafer-scale computing💜 GPU FLOPS per $ doubles — every 2.5 years — Huang's Law holds🔴 Sandy Bridge (2011) — 995 million transistors — near 1 billion⚡ Nvidia Rubin (2025) — 336 billion transistors — new record🔵 Intel 4004 (1971) — 2,300 transistors — first commercial CPU🟢 Apple M3 Ultra — 184 billion transistors — 80M× more than 4004⚡ Nvidia B200 (2024) — 208 billion transistors — Blackwell AI beast📉 Process node — 10,000nm → 2nm — 5,000× smaller transistors🔮 Intel 2030 target — 1 trillion transistors — angstrom era💚 AMD EPYC Rome — 39.5B transistors — chiplet revolution (2019)🔴 Moore's Law slowing — 3-year doublings now — was every 2 years💡 Pentium (1993) — 3.1 million transistors — first desktop revolution🟡 Core 2 Duo (2006) — 291 million transistors — multi-core era🔵 Apple M1 (2020) — 16 billion transistors — ARM conquered Intel⚡ Nvidia H100 (2022) — 80 billion transistors — the AI gold-rush chip🟢 TSMC 2nm — volume production 2025 — pushing physics to the limit🔵 Cerebras WSE-3 — 4 trillion transistors — wafer-scale computing💜 GPU FLOPS per $ doubles — every 2.5 years — Huang's Law holds🔴 Sandy Bridge (2011) — 995 million transistors — near 1 billion⚡ Nvidia Rubin (2025) — 336 billion transistors — new record

Moore's Law in Context

The Intel 4004 had 2,300 transistors in 1971. Nvidia's B200 packs 208 billion in 2024 — that's 90 million times more in one human lifetime.

Why log scale? On a regular (linear) scale, the Intel 4004 bar would be 0.000001% as wide as the B200 — completely invisible. A log scale makes each step represent a 10× increase, so you can see every era. The x-axis labels (1K, 1M, 1B) are each 1,000× apart.

Transistor Count by Era (Log Scale)

Each chip below represents a landmark moment. The gap from 1971 to 1993 looks similar to the gap from 2022 to 2024 — but it represents millions of times more transistors.

Source: Wikipedia Transistor Count · Intel, Apple, Nvidia chip spec sheets · Colors = era, not company

Dawn of Computing

PC Revolution

Multi-Core Era

ARM Takeover

AI Gold Rush

Peak Silicon

Blackwell Frontier

The Numbers That Changed Everything

Transistor Count, 1971–2025 (Log Scale)

A straight line on a log scale means perfect exponential growth — Moore's Law held almost exactly for 40 years. Watch the line steepen after 2018 as Nvidia's AI chips push count far above the historical trend. Dot color shows which company set the record at each point.

Source: Wikipedia Transistor Count · Intel, AMD, Apple, Nvidia chip spec sheets

Intel

AMD

Apple

Nvidia

Process Node: How Small Are Transistors?

A "node" is the approximate size of a transistor's features. Going from 10,000nm (1971) to 2nm (2024) means transistors shrank 5,000×. Lower = smaller = faster and more efficient. Note: modern "nm" numbers are marketing labels, not literal sizes.

Source: anysilicon.com · Wikipedia 2nm/3nm process nodes

Huang's Law: GPU Performance per Dollar

Named after Nvidia CEO Jensen Huang — GPU compute per dollar doubles roughly every 2.5 years. GFLOPS = billions of calculations per second. In 2006: 0.5 GFLOPS/$. Today: 2,100 GFLOPS/$. That's a 4,200× improvement — enabling AI that was physically impossible 20 years ago.

Source: epoch.ai/blog/trends-in-gpu-price-performance

Is Moore's Law Slowing Down?

Moore originally predicted a doubling every year (1965). He revised it to every 2 years in 1975 — and that held remarkably well for 4 decades. Now it's closer to 3 years per doubling, and getting harder.

How to read this: Green = on or ahead of Moore's 2-year target. Orange = slightly behind. Red = significantly slowing. Intel CEO Pat Gelsinger publicly acknowledged in 2023: "we're doubling effectively closer to every three years now."

Source: Calculated from transistorTimeline data · Pat Gelsinger, CES 2023 · Epoch AI research

Today's Leaders: Who Packs the Most Transistors?

These are the densest chips shipping today. Nvidia leads because AI training demands massive parallelism — the H100 has 80 billion transistors, all designed to run matrix math for neural networks simultaneously. Apple's M-series chips are CPUs that compete in absolute transistor count by fusing CPU + GPU + memory controller into one package.

Source: Nvidia, Apple, AMD, Intel chip spec sheets 2024–2025

Intel

AMD

Apple

Nvidia

Key Milestones, 1971–2030

Every crossing of a significant transistor threshold. Each one unlocked something that previously seemed impossible.

Year	Chip / Event	Transistors	Process	Why It Mattered
2030	1 Trillion Target	1T	<1nm	Angstrom era
2025	TSMC 2nm	—	2nm	Physics limit approaches
2024	Nvidia B200	208B	4nm	2× in 2 years
2022	Nvidia H100	80B	4nm	AI gold rush
2020	Apple M1	16B	5nm	Architecture shift
2006	Core 2 Duo	291M	65nm	Parallelism wins
1993	Intel Pentium	3.1M	800nm	Consumer computers
1989	Intel 486	1.2M	1µm	Crossed 1 million
1978	Intel 8086	29,000	3µm	Still powers your PC
1971	Intel 4004	2,300	10µm	Launched the era

The Chips That Defined Each Era

Intel 4004

Intel · 1971

2K

transistors

Intel 486

Intel · 1989

1M

transistors

Core 2 Duo

Intel · 2006

291M

transistors

Sandy Bridge

Intel · 2011

995M

transistors

AMD EPYC Rome

AMD · 2019

39.5B

transistors

Apple M1

Apple · 2020

16.0B

transistors

Nvidia H100

Nvidia · 2022

80.0B

transistors

Apple M3 Ultra

Apple · 2024

184.0B

transistors

Nvidia B200

Nvidia · 2024

208.0B

transistors

The Cracks in the Law

Chiplets Break the Monolith

AMD's EPYC Rome (2019) reached 39.5B transistors by connecting 8 separate 7nm dies — a trick that let them leapfrog Intel without a more advanced process node.

Wafer-Scale Computing

Cerebras WSE-3 packs 4 trillion transistors by using an entire silicon wafer as a single chip. Not a conventional design — but it proves the next frontier is packaging, not just lithography.

The 10nm Decade

Intel spent from 2016 to 2021 stuck on 10nm — a node so hard they renamed it "Intel 7" to mask the delay. TSMC pulled ahead and never looked back.

Moore's Law Slowing

Intel's former CEO admitted in 2023: "we're doubling effectively closer to every three years now" — not two. Physics is pushing back.

What The Numbers Actually Mean

The Doubling That Built the Modern World

Moore's Law has compounded for 54 years. A chip doubling every 2 years over 54 years is 2^27 = 134 million× more capable — a fact that explains why software eats everything.

AI Was Just Waiting for the Hardware

The math behind neural networks was known in the 1980s. What was missing was the compute. The Nvidia H100's 80B transistors made 2022's AI revolution physically possible.

The Node Name Stopped Meaning the Size

TSMC's "3nm" node has gate lengths closer to 18nm physically. The marketing name became a performance benchmark, not a measurement. Physics is harder to fake than branding.

Intel's Target: 1 Trillion by 2030

Intel has publicly stated the goal of 1 trillion transistors in a single package by 2030. At the current pace, that's achievable — but every doubling now costs more than the last.