Imagine trying to fit every single person on Earth — all 8 billion of them — inside the city of Tokyo. Shoulder to shoulder, cheek to cheek, no breathing room. Now imagine doing that twelve times over.
That’s roughly what IBM just did. With transistors.
On June 25, 2026, IBM announced the world’s first sub-1 nanometer chip — a piece of silicon the size of your fingernail that holds 100 billion transistors. For context, a human red blood cell is 7,000 nanometers wide. These new transistors are just 0.7 nanometers, or 7 angstroms. That’s about the width of a single strand of DNA.
And it’s not just a record for the history books. This chip could cut the time it takes to train a frontier AI model from three months to two weeks.
What Even Is a Transistor?
Think of a transistor as a tiny light switch. When it’s on, electricity flows. When it’s off, it doesn’t. By combining billions of these switches, your computer can do arithmetic, run apps, and stream videos — all by flipping switches billions of times per second.
The golden rule of computing for the last 60 years has been: make those switches smaller. Smaller switches mean you can fit more of them, which means more power, more speed, and less battery drain. This rule — often called Moore’s Law — has been the engine behind every smartphone, laptop, and data center on the planet.
But here’s the problem. We’re running out of room to go smaller in two dimensions.
IBM’s answer? Go taller.
Building Chips Like a City
Their new architecture, called Nanostack, treats transistors like floors in a skyscraper instead of houses spread across flat land. Instead of shrinking everything sideways, IBM stacked transistors on top of each other in three dimensions.
It sounds simple. But pulling it off requires bonding two ultra-thin silicon wafers together with almost zero defects — at a scale where even a single stray atom in the wrong place can break the whole structure. IBM’s team cracked this problem after years of research, developing a new bonding technique precise enough to work at seven angstroms.
The result is a chip that delivers 50% more performance than IBM’s previous best (their 2nm chip from 2021), or alternatively, can do the same job using 70% less power. Choose your adventure.
Why This Matters for AI
Right now, training a large AI model — the kind that powers ChatGPT or Claude — requires enormous clusters of chips running for months. Current AI accelerators can execute about 1,500 trillion operations per second (TOPS). IBM’s 7 angstrom chips could push that to 9,000 TOPS — six times more.
Put that in practical terms: a model that takes three months to train today could be done in roughly two weeks. The labs that can run more experiments, faster, win the AI race. Cheaper, faster chips are the fuel.
The new chip also dramatically improves on-chip memory (SRAM) by 40% — the biggest leap in that category in over a decade. This matters because AI models are constantly reading and writing data; slow memory is one of the biggest bottlenecks holding AI performance back.
This Is Just the Beginning
IBM is careful to note that we won’t see 7 angstrom chips in consumer devices next year. The 2nm chips IBM announced in 2021 are still rolling out commercially now. The sub-1nm technology marks where the frontier of research is — a decade ahead of what’s in your phone.
But that’s exactly the point. IBM’s roadmap shows at least a decade of future scaling using this new Nanostack architecture. The era of 2D chip shrinking may be over, but the era of stacking upward is just beginning.
There’s a certain poetry to it. Computing started with transistors big enough to hold in your palm. Now they’re smaller than the double helix that makes you human. And somehow, the engineers keep finding ways to make them smaller still — just in a new direction.
The angstrom era has begun.
This post has been created by Claude AI.
References
- IBM Debuts World’s First Sub-1 Nanometer Chip Technology — IBM Newsroom
- Introducing the first sub-1 nanometer node chip — the smallest, most powerful chip technology in the world — IBM Research
- IBM creates first sub-1nm computer chip — cramming 100 billion transistors into a tiny fingernail-sized space — Live Science
- IBM has unveiled chip technology that could help extend Moore’s Law another decade — MIT Technology Review