💴 Japan's government just put another 150 billion yen into Rapidus, the four-year-old startup trying to build the world's most advanced chips in the snow of Hokkaido. That money pushes the state's share of the company's capital to roughly 60%. Which raises an obvious question: at what point does a "private" venture quietly become a nationalized one? The answer Tokyo has engineered is stranger — and more deliberate — than the headline suggests.

The 150 billion wasn't a rescue. It was on the calendar.

It would be easy to read a near-billion-dollar government check as a panic move — a sign that Rapidus is burning cash and the state is plugging a hole. It isn't. The 150 billion yen (about $940 million at roughly 160 yen to the dollar) was written into Japan's fiscal 2026 budget months ago, with the plan to add it on top of an earlier round from April onward. Early this June, the Information-technology Promotion Agency (IPA), a body under the Ministry of Economy, Trade and Industry, actually executed it.

Stack it on the 100 billion yen the government put in back in February, and the public side has now injected 250 billion yen of equity. Together with private money, Rapidus's stated capital and capital reserve come to about 424.9 billion yen (roughly $2.65 billion). And that's only the equity slice. Counting research grants already paid — about 1.7 trillion yen — plus subsidies still to come, total government support is heading toward 2.9 trillion yen, somewhere around $18 billion, by fiscal 2027. This was a scheduled installment, not an emergency transfusion.

The state owns most of it. It still isn't a state enterprise.

On the balance sheet, yes: the government now holds around 60% of Rapidus's capital, making it the single largest shareholder. By the usual gut check — who put in the most money? — that looks like a state company.

But money and control were split apart on purpose. Most of the government's shares are a special class that carries no voting rights. Tokyo's actual voting power is capped at 11.5% — the bare minimum to be the top shareholder on paper, and far too little to run the place. The non-voting shares only flip to voting shares under one condition: if Rapidus's business deteriorates badly and talks fail to turn it around. In normal times, the government sits quietly and lets management move fast — the whole point of keeping a chip venture nimble.

On top of that, the state holds a single "golden share," a veto over a short list of grave decisions: appointing or removing directors, mergers, and the like. That's an economic-security backstop — a tripwire against the company or its technology slipping into the wrong hands — not a steering wheel.

And there's a quieter logic to taking equity rather than just handing over grants. Subsidies are gone the moment they're spent; the public never sees them again. Equity can come back. If Rapidus reaches profitability and pays dividends, the government earns a return on its ~60% stake. If the company lists on the stock market — the plan points to fiscal 2031 — the state can sell its shares and recover what it put in, potentially with a gain. Whether that actually happens depends on profits and the fine print of those share classes, but the design is built to recover, not just to spend.

It also changes how this looks from abroad. The U.S. CHIPS Act and Taiwan's incentives lean heavily on grants. Japan has layered a big equity position on top of its grants — closer to a venture investor than a pure subsidizer.

"Japan can't make 2nm." So where did that line go?

A year and a half ago, the skepticism was loud and not unreasonable. Here was a company with no recent experience below 40 nanometers announcing it would leap straight to 2nm — a node TSMC and Samsung spent years and fortunes reaching. Plenty of engineers said, flatly, that it couldn't be done.

Then the timeline started moving fast. Rapidus brought in an ASML EUV lithography machine in December 2024, confirmed it could expose and develop patterns by April 2025, installed over 200 single-wafer tools by June, and on July 18, 2025 confirmed that 2nm GAA (gate-all-around) transistors made on its IIM-1 line in Chitose actually worked. Chairman Tetsuro Higashi called the pace unusual even by world standards. The transistor architecture came from IBM; the advanced-process know-how from Belgium's imec. This was never a purely "made-in-Japan" effort, and that borrowed expertise is part of why it moved quickly.

But it's worth being honest about where things stand, because Rapidus itself is. CEO Atsuyoshi Koike has described the climb to volume production as anything but easy: pushing yields — the share of good chips per wafer — up to the 60–70% range needed to make money is, in his words, not a simple world. A pilot line is running, but as one industry writer put it, the climb has barely reached the first station of the mountain. Making one transistor work is a milestone. Making tens of thousands of wafers a month at a sellable yield is a different problem entirely.

The roadmap: PDKs (the design kits customers need) to lead clients during fiscal 2025, volume production starting in the second half of fiscal 2027 at about 6,000 wafers a month, then a roughly fourfold ramp to 25,000 a month in fiscal 2028. The node is called 2HP, with a logic density reported on par with TSMC's N2.

Building the chips is half the battle. Selling them is the other half.

This is the question that keeps engineers up at night more than yield does. A fab with no steady stream of orders can't improve its yields or justify its running costs, subsidies or not. Customer acquisition is widely seen as the single biggest risk in the whole Rapidus story.

The early signs are real but still early. In March 2026, Canon became the first major Japanese end-user to sign on as a prospective customer, ordering prototype image-processing chips. Fujitsu has talked about using Rapidus to make cutting-edge AI processors — NPUs — around 2029. Rapidus is co-developing CPUs with U.S. startup Tenstorrent, and Tokyo's Preferred Networks is weighing whether to fab its AI chips there. Koike has said the company is in talks with more than 60 potential customers.

To make sure there's demand from day one, METI's affiliate NEDO has tied Rapidus together with Fujitsu and IBM Japan in commissioned projects — a deliberate effort to manufacture anchor customers rather than wait for them. And Rapidus's pitch isn't to out-volume TSMC. It's the opposite: small batches, fast turnaround, for designers who want a foundry that isn't concentrated in Taiwan. In a world nervous about single-region risk, "the other option" is a real product.

Two years behind at the gate — is there still a lane to win?

Rapidus is starting the race a couple of laps down. By late 2025, all three incumbents had moved into 2nm-class production: Samsung with SF2, Intel with 18A, and TSMC with N2 at two fabs in Taiwan. TSMC alone controls roughly 70% of the foundry market and is the only one really supplying leading-edge nodes to outside customers at scale; Intel's and Samsung's external foundry businesses remain comparatively thin. When Rapidus reaches volume in fiscal 2027, it'll be roughly two years behind.

But the AI boom has done something unusual: demand for top-tier nodes has outrun even TSMC's enormous capacity. That gap — the "opening window" the industry keeps talking about — is exactly what a latecomer needs. You don't have to beat TSMC to be useful; you have to exist as a credible second source when the leader is sold out and customers are spooked by geography.

That's also where the geopolitics and the funding structure rejoin. Japan isn't only chasing chips; it's buying optionality in a supply chain that runs through one of the most contested straits on earth. The unusual equity-heavy model — state as majority owner, private sector as the steering hand, a golden share as the emergency brake — is Tokyo's attempt to fund that bet at national scale without smothering the company that has to win it.

Whether it pays off comes down to two numbers nobody can promise yet: the yield on a wafer, and the length of the customer list. Everything else is scaffolding around those two.

Japan has decided this is worth tens of billions of dollars and a deliberately tangled ownership structure to attempt. In your country, how far should the state go — and in what form — to keep a critical industry at home?

References