TSMC's first Arizona fab doubled from $12B to $20B — most of the '$165B blowout' is other fabs

In May 2020, a governor stood up in Arizona and announced that the most advanced chipmaker on earth would build a $12 billion factory in the desert, with chips rolling off the line by 2024.¹² Five years later, the headline number attached to that same patch of Phoenix desert was $165 billion — a figure so large it gets narrated as one of the great construction disasters of the decade, a $12 billion project that supposedly ballooned fourteenfold.¹¹ That story is wrong in the most interesting way possible: almost none of the growth is overrun. It's a company deciding, four separate times, to build more.

The official retelling is that TSMC's Arizona bet spiraled out of control: $12 billion, then $40 billion, then $65 billion, then $165 billion, a runaway budget nobody could stop. The real story is a moving target, not a busted one. Each of those jumps is a new commitment — a second fab, a third fab, a fourth expansion — layered on as CHIPS Act money and AI demand showed up. The genuine overrun is real, and it's much smaller and much more telling than the number people are angry about.

Four commitments wearing one price tag: the leap from $12B to $165B is scope stacked on scope, not a budget torn to pieces

Walk the sequence and the confusion dissolves. In 2020 TSMC committed to one fab at about $12 billion.¹ In December 2022 it disclosed it had started a second fab alongside the first, taking the combined investment to roughly $40 billion — described in its own SEC filing as the largest foreign direct investment in Arizona history.³ In April 2024, with a CHIPS Act preliminary agreement in hand, it announced a third leading-edge fab, pushing total planned capital expenditure at the Phoenix site above $65 billion.⁶ Then in March 2025 it announced an additional $100 billion, bringing the total to $165 billion.¹¹ Every one of those figures is the cumulative total of a bigger project, not the ballooning cost of the original one. The number grew because the ambition grew — three greenfield fabs by late 2024, and more coming.⁷

Here is the thesis in one line: the eye-popping growth from $12 billion to $165 billion is mostly added fabs, not overrun on the first one. The only place a real, verifiable overrun lives is inside that first fab — and it's a fraction of the drama the big number suggests.

The overrun that actually happened: one fab doubled from $12B to $20B and slipped about a year — that's the whole verifiable damage

Strip away the second, third, and fourth commitments and look only at the original fab, the one announced in 2020. By late 2024, that single fab's price tag had risen from about $12 billion to roughly $20 billion, and its full-production date had moved from 2024 to 2025.⁹ That's a real overrun — on the order of two-thirds above the original single-fab budget — and it is the honest, specific damage the headlines should be reporting. It is also a rounding error next to the $165 billion figure it keeps getting fused with. A doubling is not a fourteen-bagger. The confusion isn't accidental; a $165 billion 'blowout' is a far better story than a fab that cost $8 billion more than planned.

Why the fab was late: not COVID, but a labor shortage: the delay wasn't broken supply chains — it was that the desert didn't have the technicians the fab needed

The 'years late' framing overstates it, and the usual cause — 'COVID and supply chains' — misses the point. On the first fab specifically, TSMC's December 2022 filing still targeted 2024 for N4 production.³ The slip disclosed in mid-2023 pushed that to 2025: about a one-year delay, not multiple years.⁴ And the cause TSMC's own leadership pointed to wasn't a shipping crisis. It was a shortage of skilled local labor — serious enough that the company flew technicians in from Taiwan to train Arizona hires.⁴ Pandemic surges and licensing issues were secondary. This is the mechanism worth sitting with: the hardest thing to import into the Arizona desert wasn't machines or capital. It was the specialized workforce that runs a leading-edge fab. You can pour concrete on a schedule. You cannot conjure fab technicians on one.

The delayed, over-budget fab that beat Taiwan: the part the disaster narrative skips: once it ran, the numbers were embarrassingly good

The disaster framing has a problem: the fab worked, and worked well. In October 2024, TSMC's U.S. division president said the Arizona fab was hitting production yields about four percentage points higher than comparable facilities in Taiwan.⁸ By a November 2024 site visit it was in pilot production, making sample wafers for customer verification — reportedly already sending chips to customers including Apple.⁹ In January 2025, the U.S. Commerce Secretary confirmed the fab had begun producing 4nm-class chips, the first time that cutting-edge node had been manufactured on American soil, with yield and quality on par with Taiwan.¹⁰ A project that overran by a third and slipped a year still delivered leading-edge chips at world-class yields. That is not a construction failure. It is a hard build, done.

Isn't a doubling still a failure?: the honest objection: a ~67% overrun and a year late is nothing to celebrate — and it isn't nothing

The fair objection is that even the real number is bad. A fab that costs $20 billion against a $12 billion plan and arrives a year late is a genuine overrun, and rooting it in a labor shortage doesn't erase it — TSMC chose the site knowing the workforce would be thin. That's true, and it's the honest counter. But two things blunt it. First, greenfield leading-edge fabs are among the most complex facilities humans build, and doing a first one in an unfamiliar labor market, in a new country, at world-class yields is a different kind of feat than repeating one in Taiwan. Second — and this is the tell — TSMC didn't retreat after the overrun. It committed another $100 billion after the first fab was substantially complete.¹¹¹² Companies don't quadruple down on a project they believe failed. The overrun was the tuition. The four commitments are what the tuition bought: a workforce, a supply base, and a running leading-edge fab on U.S. soil — the vertical footprint the whole exercise was for.

TSMC's Arizona 'blowout' is a lesson in how a number becomes a myth. Say '$12 billion to $165 billion' fast enough and it sounds like a company that lost control of a construction site. Slow it down and it's a company that kept deciding to build more, on the strength of a first fab that overran by a third, slipped a year, and then out-yielded its own factories in Taiwan.⁸ The desert didn't defeat the plan. It just charged more than expected for the one thing that couldn't be shipped in a crate — the people who know how to run the machines.