AINEWS
TSMC is adding $100 billion for Arizona chip factories and 2 nanometer production. The advanced packaging step that turns chips into AI accelerators is still scaling in Taiwan.
TSMC will spend another $100 billion on its Arizona chipmaking complex, the company said on its second-quarter earnings call, lifting its cumulative U.S. commitment to $265 billion across up to 12 fabrication plants. The new money, framed by chief executive C.C. Wei as "additive" rather than a relocation, targets the most advanced node TSMC makes: 2-nanometer transistors, the smallest geometry in commercial production, which began shipping from Arizona in the second quarter. The 2nm production start is the structurally significant piece of the announcement. The dollar figure is the headline.
On the same call, TSMC also committed to 13 new leading-edge and advanced packaging fabs in Taiwan over the next several years. That parallel commitment, made on the same day, is what "additive" actually means in this industry. The Arizona build is not pulling capacity out of Hsinchu and Tainan. It is adding to both sides at once, at a scale neither side has tried before.
Advanced packaging is the part of the chipmaking chain most readers will not recognize, and the part that decides whether the U.S. "supply-chain resilience" story holds up. A 2-nanometer transistor is a feature on a silicon wafer. It is not a working chip. Turning a finished wafer into a working AI accelerator requires advanced packaging: stacking, chiplet integration, and the high-bandwidth memory interfaces that connect a GPU's compute tiles to the memory that feeds them. NVIDIA's Blackwell and Rubin accelerators, AMD's MI series, and Broadcom's custom AI silicon all depend on TSMC's CoWoS (chip-on-wafer-on-substrate) advanced packaging, and the bulk of that capacity still lives in Taiwan. The new Arizona money includes advanced packaging fabs, but the volume ramp is years behind the Taiwan line.
The earnings backdrop explains why TSMC can write the check. The company reported a 77% year-over-year profit surge in the second quarter to a record roughly $22 billion, on revenue driven by AI accelerator demand from NVIDIA, AMD, and Broadcom. With a market capitalization near $2.18 trillion, TSMC is the most valuable foundry in the world by a wide margin, and the only one operating at the 2nm node in volume. That concentration of profit and process leadership is the financial engine under the $265 billion figure, not a gesture of corporate patriotism.
policy frame is part of why the announcement reads the way it does. The CHIPS and Science Act of 2022 set the conditions for federal subsidies and tax credits that effectively underwrite a meaningful share of the Arizona capex. The Trump administration's tariff regime, including the semiconductor-specific tariffs threatened in 2025, has added a cost to running advanced nodes exclusively in Taiwan that did not exist two years ago. Arizona is the answer both policies wanted. Whether the answer reduces Taiwan dependence is a different question, and the 13 Taiwan fabs on the same earnings call suggest TSMC's own read is that it does not.
Three lanes are worth watching. First, 2nm yield parity: whether the Arizona line can hit the same defect density as the Taiwan fabs on the same node, on the same timeline. Second, advanced packaging volume in Arizona: when the planned U.S. CoWoS-equivalent capacity actually ramps, and at what fraction of Taiwan's output. Third, the Taiwan fabs themselves: whether the 13 are commissioned in series as planned, or whether capital is quietly redirected toward Arizona if the U.S. subsidies and tariff economics make Arizona more attractive than the announced cadence implies.
The earlier Arizona track record is what makes those three lanes worth tracking. The original Phoenix fab, announced in 2020, slipped years past its initial production target and ran into labor disputes during construction. The current Arizona workforce is unionized under a different framework than the one originally agreed, and water access in the Phoenix metro has been a recurring planning constraint that state regulators have had to address at each expansion. A $265 billion buildout, even spread over a decade, will run through those same constraints, and the question of how many of the promised high-paying jobs materialize at the rate TSMC has previously delivered is open.
TSMC is building out the most advanced node it has, on U.S. soil, in volume, at the same time it is scaling the same node and the advanced packaging that surrounds it in Taiwan. The dollar figure is real industrial commitment, not a press-release gesture. The 13 Taiwan fabs on the same earnings call are the reason that commitment is also, plainly, not the whole story.