TSMC Arizona turned a profit faster than anyone predicted. That is the headline TSMC wanted. The more load-bearing data point is what its own chairman told Taiwan's National Development Council minister after a site visit: four structural constraints that do not go away when you build a second fab.
Water is a physical limit. TSMC's Arizona site sits in one of the driest metro areas in the U.S., and semiconductor fabs are among the most water-intensive industrial facilities that exist. Visa processing for Taiwanese process engineers is a U.S. policy variable that has not improved. State utility regulation in Arizona has added friction to power interconnection timelines. Power supply uncertainty adds schedule risk to the two fabs already under construction and to any future capacity expansion. None of these are solvable by semiconductor policy alone. The second fab finished construction with production scheduled for H2 2027; the third fab broke ground. Each additional fab adds capacity and amplifies the dependency on unresolved infrastructure.
The $514M profit number is real and it validates the yield and process transfer. It also makes the constraint list harder to dismiss as new-fab growing pains. Teams designing western supply-chain redundancy into their AI hardware programs should treat Arizona's water, power, and labor constraints as multi-year planning inputs, not temporary friction. Intel's Ohio and Germany fabs face similar or worse versions of the same problems. The geography that avoids geopolitical risk creates infrastructure risk.