TAIPEI – TSMC CEO Che-Chia Wei told analysts Thursday that AI chip demand is “extremely robust” and expected to remain “very strong until around 2029 or 2030.” That is not a forecast hedge. It is a four-year order-book signal from the company that manufactures the processors inside nearly every major AI system on earth, and it landed alongside the numbers that explain why TSMC is in a position to say it with confidence.
Taiwan Semiconductor Manufacturing Company posted second-quarter net profit of NT$706.6 billion, equivalent to roughly €19.1 billion, a 77 percent increase over the same quarter last year. Revenue reached $40.2 billion, up 36 percent, beating analyst estimates of $39.63 billion. Euronews reported the results Thursday alongside TSMC’s announcement of a further $100 billion committed to US manufacturing expansion.
The company raised its full-year revenue growth forecast to slightly above 40 percent year-on-year, up from previous guidance of 30 percent or more. June revenue alone jumped 67.9 percent compared with June 2025, and first-half sales rose 35.6 percent. Alongside that, TSMC increased its capital expenditure budget for the year to between $60 billion and $64 billion, compared with the prior range of $52 billion to $56 billion. That is a significant upward revision for a company that has been spending at levels most semiconductor competitors cannot approach.
The new $100 billion US investment, announced alongside the earnings results, will fund four additional fabrication plants in Arizona dedicated to producing 2-nanometer and more advanced chips. They add to the six Arizona fabs TSMC had already committed to building under a prior $165 billion pledge, bringing the total US manufacturing commitment to approximately $265 billion. That is a number that existed nowhere in TSMC’s investment plans two years ago, and it reflects how quickly AI demand has redrawn the company’s capital allocation at the highest level.
The AI demand driving these results is not a single product line. TSMC manufactures the advanced chips used in data center accelerators by Nvidia, AMD, and other AI infrastructure companies, the application processors in Apple’s iPhone and Mac product lines, and an expanding set of custom silicon designs from hyperscalers including Google and Amazon. When Wei says demand is “extremely robust,” the claim is grounded in a customer base that has collectively committed to purchasing levels that require TSMC to run its most advanced nodes at full capacity for years.
The $265 billion US commitment, while presented as an investment, is also a political calculation. The CHIPS and Science Act created subsidy incentives for advanced semiconductor manufacturing on American soil, and the Trump administration has applied additional pressure on Taiwan to direct more production to the United States. TSMC’s cumulative US pledge is, at least in part, a response to those incentives and that pressure. The question of whether it is also a smart business decision depends on whether Arizona can become the kind of place where advanced chip fabrication actually works at scale.
That question remains genuinely open. TSMC’s Taiwanese fabs benefit from decades of accumulated engineering culture, a dense supplier ecosystem, and a workforce trained in the specific disciplines that advanced chipmaking requires. The first Arizona fab, TSMC’s N4 plant producing 4-nanometer chips, has faced documented production challenges and lower yield rates than comparable Taiwan facilities, partly attributed to workforce training gaps. Scaling that to ten total Arizona fabs producing 2-nanometer and below chips represents a different challenge from the initial investment.
That challenge is the context against which Huawei’s competing approach matters. China’s chip design ecosystem has aligned behind Huawei’s Tau Scaling Law, a framework for reaching 1.4-nanometer-equivalent transistor density by 2031 without ASML’s extreme ultraviolet lithography equipment, which export controls have blocked China from acquiring. If that approach succeeds, the United States and Taiwan will face a China capable of matching advanced semiconductor performance through a parallel technical path. The $265 billion US expansion is, in that context, also a hedge against that possibility.
TSMC’s earnings beat what Wall Street expected, and its guidance implies the company believes the AI buildout will sustain demand well past what most analysts were projecting at the start of 2025. The semiconductor cycle has historically been volatile, with boom periods followed by inventory gluts and demand corrections. The current AI infrastructure investment wave is large enough that some analysts describe it as structurally different from prior cycles, driven by corporate and government spending commitments that have longer time horizons than consumer electronics demand. Others caution that the AI buildout will eventually saturate, and that TSMC’s capex commitments will look different once that happens.
Wei’s 2029-2030 AI demand outlook is more than a year further than most public analyst forecasts go. The specific reasons TSMC’s CEO has longer visibility than external observers are not publicly explained: TSMC does not disclose individual customer order books, and the company’s guidance has historically been conservative rather than promotional. Those who believe the forecast take it at face value; those who do not have to argue that the world’s most important chip foundry is wrong about its own order pipeline.
The four new Arizona plants announced Thursday will not be operational before the early 2030s. A 2-nanometer fab does not get built and qualified in 18 months. The investment is real, the demand thesis behind it is visible in the quarterly numbers, and the workforce and supply chain questions are not fully answered. TSMC is committing $265 billion to a bet that the AI chip cycle lasts long enough, and that Arizona can become capable enough, for that figure to make sense. Thursday’s results suggest the first half of that bet is currently winning.
