For years, Japan treated China the way most businesses treat their biggest supplier: reliable, efficient, slightly uncomfortable, but too convenient to question.
Then the cracks started showing.
Tourism pressure. Military standoffs in the airspace. Export controls creeping into sensitive materials.
The gentle reminder that trade can turn political very quickly when someone feels like pulling a lever.
Japan did what it always does: tightened its belt and started planning ten years ahead.
It had already learned this lesson once. Back in 2010, a rare-earth export freeze from China sent Tokyo scrambling.
Since then, Japan quietly diversified suppliers, built stockpiles, and pushed manufacturers to spread production beyond China under the “China Plus One” strategy.
Then Covid delivered the real punch.
Car factories stalled, and electronics vanished off shelves.
Suddenly the world discovered that the tiny silicon chips inside everything – from cars to washing machines to servers – weren’t just industrial components anymore.
They were strategic assets.
And just as the supply chain was wobbling, artificial intelligence showed up demanding an industrial-scale power bill.
Data centres are multiplying, power consumption is climbing, and the entire digital economy now depends on whether chips can keep getting faster without melting the grid.
That’s where Japan realised that it no longer controlled the heart of modern computing.
The gap Japan left behind
Japan used to dominate semiconductors.
Then the industry moved fast, expensive, and brutally global. Advanced logic manufacturing marched on.
Japan stopped advancing around the 40-nanometre generation while Taiwan, South Korea and the US pushed into ever-smaller, ever-faster chips.
Once you fall behind in semiconductors, catching up isn’t like upgrading a factory line. It’s like trying to rejoin Formula One after skipping ten seasons.
In 2022, a group of Japanese semiconductor veterans decided Japan couldn’t afford to sit on the sidelines anymore.
They founded Rapidus.
Not as a nostalgia project, but as a leapfrog attempt straight into the most advanced manufacturing on earth: 2-nanometre logic chips using next-generation transistor structures.
If that sounds ambitious, it’s because it is.
Why smaller really does matter
A nanometre is one-billionth of a metre.
At 2nm, you’re working at scales measured in atoms. The reason the industry keeps shrinking transistors is pure physics and economics.
Smaller transistors switch faster and waste less power. Move from today’s mainstream processes to 2nm and you can reduce power consumption for the same workload by roughly three quarters, while lifting performance at the same time.
That matters when AI workloads are pushing data centres toward power limits that utilities weren’t designed for.
Faster chips that sip power instead of gulping it aren’t optional anymore. They’re survival tools.
Rapidus’ bet is that Japan can re-enter the chip race not by being bigger, but by being faster, cleaner and more integrated.
Building a factory that thinks like software
Rapidus is building its flagship manufacturing site called IIM, short for Innovative Integration for Manufacturing, in Chitose, Hokkaido.
The pilot line began operating in 2025, with mass production targeted for 2027.
But the factory itself is only half the story.
Traditional chipmaking is slow because the industry is fragmented.
Designers create chips. Foundries manufacture them, packaging houses assemble them. Data flows slowly between each step, and every iteration burns time and money.
Rapidus wants to collapse that entire loop.
Its operating model is called RUMS: Rapid and Unified Manufacturing Service.
The idea is simple in theory and brutal in execution: integrate design, manufacturing and packaging so tightly that learning cycles compress dramatically.
Designs reach silicon faster, manufacturing feedback loops directly into design, yield improves earlier, and ultimately…customers move quicker.
One of the clever tricks enabling this is single-wafer processing.
Instead of pushing large batches through the factory, Rapidus processes wafers one at a time, capturing richer data from every step.
That data feeds back into future designs and process tuning, accelerating the ramp to stable production.
Speed becomes the product.
Letting AI design the chips as well
Rapidus is also leaning heavily into AI-driven chip design through a platform called Raads – Rapidus AI-Agentic Design Solution.
Starting in 2026, customers will be able to use AI tools that translate specifications into chip designs, predict power and performance early, and flag problems before expensive mistakes get baked into silicon.
The company estimates this can cut design time by half, and reduce design costs by roughly 30%.
More importantly, Raads connects directly with real manufacturing data coming off the factory floor.
In semiconductor language, this is called Design-Manufacturing Co-Optimization. In plain English, it means fewer nasty surprises.
When it stopped being theoretical
In late 2024, Rapidus installed Japan’s first mass-production-capable EUV lithography machine, enabling true 2nm manufacturing.
By mid-2025, prototype 2nm gate-all-around transistors were already demonstrating working electrical performance.
Rapidus says it has raised about US$731 million, employs roughly 1,000 people, and has both government and major industrial backers as it moves toward mass production.
The company isn’t just building a factory.
It’s testing whether Japan can regain control over the digital engine of the modern economy.
This article is not financial advice. Always do your own research or speak with a licensed adviser before making investment decisions
Read more: BroadBand Tower: Japan’s small cap plugged into the AI boom
