Rapidus Is Talking to 60 Companies and Has Handed Design Kits to About 10 — Why Winning Customers Is Harder Than Building the Fab

The full picture of government funding and the production plan is covered in Japan Has Committed ¥2.35 Trillion to Rapidus. Here we focus on neither money nor technology, but on how to gather customers and an ecosystem.

What is a PDK, and why does handing one over matter?

A foundry is a fab that manufactures semiconductors designed by other companies. What often gets overlooked: to design for a fab, a customer needs that fab's specific way of drawing blueprints. The package that captures it is the PDK (Process Design Kit).

A PDK contains transistor characteristics data, design rules such as wiring spacing, and component libraries for verification. Only with these can a designer draw circuits for that fab's manufacturing process. Put the other way around, handing over a PDK is the act of opening the door to a customer: design on our 2nm process. The fact that Rapidus has moved to providing PDKs to about 10 companies shows customer acquisition has advanced from talks to a phase where customers can actually try designing.

How far along is Rapidus in gathering customers?

According to Rapidus president Atsuyoshi Koike, the company is in talks with more than 60 prospective customers, mainly in HPC (high-performance computing), AI and edge, and provides PDKs to about 10. The meaning of these numbers needs precision: neither 60 nor 10 is a count of signed contracts. They are companies in discussion, or companies that have advanced to trial design.

The movement still matters, because a foundry business grows into volume orders only through the accumulation of exactly this kind of groundwork. The government is also backing front-end and back-end technology development toward 2nm volume production in the second half of fiscal 2027, which lets Rapidus bring technical and financial credibility to the negotiating table.

Why are customers harder than the fab?

A leading-edge-node fab costs trillions of yen just to build. But without customers designing and manufacturing there, utilization never rises and the enormous investment cannot be recouped. This is the structural difficulty facing any new foundry.

Chip design depends on an ecosystem: EDA (electronic design automation) tools for drawing circuits, and IP (design assets) — pre-built functional blocks that designers reuse. TSMC, the world's largest foundry, spent decades building this environment and locking in a vast base of customers and design assets. A new entrant like Rapidus must earn customer trust in its design environment from zero. A fab (hardware) can be built with money; a design ecosystem (software and trust) cannot be stacked up overnight. That is why customers are harder than fabs.

How is this different from TSMC? The high-mix, short-TAT bet

The Rapidus strategy differs from TSMC's model of driving unit prices down through mass production. What Rapidus champions is high-mix low-volume production with short TAT (Turn Around Time — the period from design to prototype). By drastically compressing design-to-prototype schedules and cycling many product types quickly in small lots, it aims to capture the diverse AI and edge demand that one-size-fits-all mass production struggles to serve.

Whether this bet reaches profitability will be tested from the second half of fiscal 2027 onward, when volume-production yield (the share of good chips) and customer count must both materialize. The 2nm GAA-transistor prototyping is progressing, but commercial viability will be decided over the next few years.

For the full picture of the company, see Rapidus Explained: Japan's Last Bet on a 2nm Foundry. What the label 2nm actually means as a node name is covered in 2nm no longer means a length of 2 nanometers (in Japanese), and the details of state funding and the production plan are in Japan Has Committed ¥2.35 Trillion to Rapidus.

Key takeaways

• What decides a foundry's fate is customers, not the fab. Rapidus is in talks with more than 60 companies and now provides PDKs to about 10 (media-reported).
• A PDK (process design kit) is the specification set needed to design on a fab's process — handing it over is the practical entry point of customer acquisition.
• Chip design needs an ecosystem of EDA (design tools) and IP (design assets), which TSMC spent decades building.
• Rapidus is betting on high-mix low-volume, short-TAT production rather than mass production, targeting diverse AI and edge demand.
• Whether the strategy pays off will be tested from fiscal 2027 onward, when volume yield, customer count and timing all come together.