TSMC Built Fabs in Arizona and Japan to De-Risk Taiwan. It Kept the Crown Jewels at Home.

In the early days of building Fab 21 in the Arizona desert, TSMC ran into a problem so mundane it borders on comic: sulfuric acid, a basic plant chemical, cost roughly five times more locally than at home, so the company shipped it across the Pacific from Taiwan.⁸ Sit with that. The most sophisticated manufacturer on earth, planting a flag in the United States to prove it could make chips far from Taiwan, found it cheaper to send a tanker of acid 7,000 miles than to buy it down the road. That single detail tells you what the headlines won't: building outside Taiwan is not just harder. It is structurally, expensively, slower-by-design harder — and TSMC knows it better than anyone.

The official story is that TSMC is de-risking. After years of warnings that the world's advanced chips are made on one island within missile range of China, the company is spending up to $165 billion in Arizona and over $20 billion in Japan to spread the crown jewels around. That is the story. Here is what's actually happening: the crown jewels are not moving. They are staying exactly where they have always been, and TSMC has told us so on the record.

The thesis is simple. TSMC's overseas fabs are not a genuine de-risking of Taiwan — they are a politically-driven tax TSMC is paying to keep its biggest customers and host governments happy, while the cutting edge stays home. The proof is in the roadmap: A16, TSMC's 1.6nm-class node and its current bleeding edge, is set for volume production in 2027 — in Taiwan first.⁷ The desert and Kumamoto get yesterday's and last year's technology. Taiwan keeps tomorrow's.

Watch the number grow, and notice what it never buys

The Arizona commitment did not arrive as a plan. It arrived as a sequence of escalating press releases, each one larger than the last and each one timed to a political moment. The original 2021 SEC filing committed roughly $12 billion to a single fab, targeting commercial production in 2024.¹ By December 2022 it was two fabs and about $40 billion.² In 2024 it became three fabs and $65 billion. Then in March 2025, CEO C.C. Wei stood up and announced an additional $100 billion — six fabs, two packaging plants, and an R&D center, for $165 billion total.⁴ A fourteen-fold increase in four years is not the unfolding of a strategy. It is a company being asked, repeatedly, for more — and saying yes.

And the money came with a leash. TSMC's $6.6 billion CHIPS Act award was not a single event but a courtship: a preliminary, non-binding memorandum of terms signed in April 2024, and a binding final agreement not executed until November 2024 — seven months later — with up to $5 billion in loans attached.³ The subsidy is real, but so is the dependency it creates. You do not build a $165 billion footprint on the strength of a $6.6 billion grant unless the grant is a symbol of something larger: a government relationship you cannot afford to disappoint.

Why the cutting edge can't follow the money

Here is the mechanism the de-risking story ignores. A leading-edge fab is not a building you fill with machines — it is the dense, irreplaceable cluster of suppliers, engineers, permitting speed, and tacit know-how that surrounds it. In Taiwan, that cluster has been compounding for decades, and it shows up in the clock. TSMC's CEO has said permit approvals in the US take at least twice as long as in Taiwan.⁸ Independent engineering data from the firm Exyte puts total US fab construction at around 38 months versus 19 months in Taiwan — roughly double.⁸ When the very edge of the roadmap moves a node every year or two, a fab that takes twice as long to build and costs more to run is, by definition, the wrong place to debut your newest process. You debut it where you can move fastest. That is Taiwan, and the geography is doing the deciding.

Now look at what the overseas fabs actually make. In Japan, JASM's Kumamoto Fab 1 — opened in early 2024 — runs on 12/16nm and 22/28nm nodes: mature-to-specialty technology, generations behind the leading edge.⁵ Its second fab, where construction began in late 2025 with about $13.9 billion of investment, targets 6nm with production expected by the end of 2027.⁶ That is respectable, but it is not the edge — and tellingly, JASM is not even wholly TSMC's. The company holds roughly 86.5%, with Sony, DENSO, and Toyota holding the rest,⁵ because the original logic of Kumamoto was never geopolitical insurance. It was proximity to Sony's image-sensor demand. The de-risking narrative was draped over a customer-supply decision after the fact.

The node that proves where TSMC's heart still is

A16 is the tell. It is TSMC's entry into the Angstrom era — a 1.6nm-class node built on nanosheet GAAFET transistors and Super Power Rail backside power delivery, promising an 8–10% speed gain or 15–20% power reduction over the prior generation, aimed squarely at AI and high-performance computing.⁷ It is the product TSMC's most valuable customers will fight over. And at its April 2026 technology symposium, TSMC confirmed A16 volume production is aligned to 2027 — a slip from the previously advertised second half of 2026.⁷ The point is not the delay. The point is the location. When TSMC has its single most advanced, most demanded, most strategically precious process, it puts it in Taiwan first. The desert, with its six fabs and its $165 billion, gets to catch up later.

So the structure that emerges is not diversification. It is a hub and a set of carefully chosen satellites. Taiwan keeps the edge, the R&D, the speed, and the supplier density. Arizona and Japan absorb the political pressure and serve the customers who need US-soil or Japan-soil supply — at a node that is deliberately a step or two behind. The concentration risk everyone worried about is barely dented, because the thing that matters most, the leading edge, never moved.

But isn't a slow, expensive hedge still a hedge?

The fair objection is that this is too cynical. Capacity is capacity. If Taiwan went dark tomorrow, a fleet of running fabs in Arizona and Japan — even on older nodes — would matter enormously to the millions of products that don't need 1.6nm. Mature and near-leading-edge chips run cars, sensors, and most of the world's electronics. So the overseas fabs are insurance, just not insurance on the most glamorous policy line. That's true, and it's the strongest version of the case. But notice what it concedes: the insurance covers the commodity, not the crown jewels. The single most valuable, hardest-to-replace capability — the leading edge that the entire AI economy now depends on — remains a single-point-of-failure on one island, by design, because moving it is too slow and too costly to make sense while Taiwan holds. TSMC has hedged the part of its business it could afford to lose, and kept at home the part it cannot. That is not de-risking the core. It is de-risking the periphery and calling it the core.

TSMC is doing exactly what a rational monopolist of the leading edge should do when the world's most powerful governments ask it to move: it builds, it smiles, it accepts the grant, and it ships the acid from Taiwan. The fabs in Arizona and Kumamoto are not a bluff — they will make real chips for real customers. But they are the price of keeping everyone happy, not the relocation of what makes TSMC irreplaceable. The crown jewels stay home, behind the supplier cluster no government can subsidize into existence overnight. The world wanted TSMC to de-risk Taiwan. What it got was TSMC making Taiwan, quietly, more indispensable than ever.