TSMC completes its second Arizona fab as Intel slides its Ohio One project toward 2030

The CHIPS Act was sold as a single industrial programme. It has resolved into two quite different stories. TSMC closed construction on its second Arizona fab in April and broke ground on a third on the same Phoenix site the same month, with equipment move-in for Fab 21 Phase 2 scheduled for the third quarter and high-volume 3-nanometer production targeted for 2027. Intel, meanwhile, has formally conceded that its Ohio One campus in New Albany will not reach first-wafer completion until 2030, with high-volume production pushed to 2031 — a slip of five years from the 2026 opening the groundbreaking ceremony advertised.

The divergence is not primarily about subsidy. Both projects received CHIPS Act awards structured on similar terms. The divergence is about execution. TSMC brought Taiwan construction crews, Taiwan process engineers, and a supply chain of equipment vendors to Phoenix, and applied a manufacturing discipline the rest of the Western industry has not previously been able to replicate on US soil. Intel's difficulties are partly macro — a demand air-pocket in foundry services as its Gaudi and CPU franchises have lagged — and partly operational. Ohio was a future-fab bet predicated on 18A yield. The yield ramp has not arrived on the schedule the site plan required.

The specific numbers are unflattering. TSMC Arizona's first fab, according to the company, is now producing on 4-nanometer at US yields the company has publicly characterised as comparable to its Taiwan operations. Nvidia Blackwell dies are being assembled from those wafers, as are parts of Apple's A-series and M-series portfolio. Intel Ohio has no comparable output profile for the remainder of the decade. The first steel on the New Albany site has not, as of early 2026, been followed by the second wave of construction its design demanded.

The winners are TSMC, whose US footprint now anchors Apple's and Nvidia's domestic-content stories, and the Arizona suppliers who have quietly built themselves into a second-source ecosystem. The losers are Ohio's state government, whose economic-development assumptions were front-loaded, and the US Commerce Department, whose narrative about broad geographic diversification of domestic advanced logic capacity is harder to tell with one working site.

What the divergence forecloses is the assumption that subsidy flows mechanically into capacity. What it opens is the uncomfortable question, not much discussed during the CHIPS negotiation, of whether the United States has more than one fabless-era company capable of building and running a leading-edge fab inside its own borders on schedule.