TSMC pumps another $100 billion into Arizona, and the real story is packaging

Those of you who follow probably know that TSMC pumped another 100 billion dollars into Arizona. A closer look at this news shows the problem

100 billion dollars, and a question that everyone ignores: the specific timeline, where does the number 4 factories come from?

On July 16, 2026, at the announcement of Q2 business results in Taipei, TSMC CEO C.C. Wei announced a number that made both technology and investors pay attention: an additional $100 billion will be poured into Arizona to build at least four new chip factories, along with advanced packaging facilities. This number brings TSMC's total investment commitment in the US to $265 billion. This confirms a market rumor that appeared in February this year and most of the press will stop there, considering this as news about "the world's largest chip company pouring more money into the US." But the more interesting question is not the number, but: where does this money actually go, and why does TSMC still refuse to give a specific timeline despite having a record business quarter?

In addition, another issue that needs to be mentioned is the number of four factories. Basically, a chip factory at the 2 nm process with a capacity of about 20,000 semiconductor wafers per month usually costs from 25 to 35 billion dollars to build and equip machinery. Divide 100 billion by that cost, and you get exactly four factories. This shows that the number "at least four" that Wei mentioned is not an emotional estimate, but almost a reverse calculation from the budget. With this new investment, TSMC's Arizona plan will reach about halfway to its ultimate goal of 12 factories and 4 packaging facilities, a scale that was mentioned in the press in April as part of a broader investment agreement between the US and Taiwan.

The real bottleneck is not the wafer factory and the geopolitical factor behind the investment check

This is the part I find most worth mentioning, and also the part most easily overlooked when skimming the news. Most of the attention is focused on wafer fab factories, but according to TSMC itself, advanced packaging capacity, a technology called CoWoS (Chip-on-Wafer-on-Substrate, meaning stacking multiple chip layers on top of each other to increase connection density and speed), is the real bottleneck in the AI ​​chip production chain, not raw wafer output. Companies like NVIDIA and AMD have long had to wait in line for CoWoS capacity more than for chips to be produced. TSMC's move to put a packaging facility right in Arizona, instead of just having a wafer factory that has to be sent back to Taiwan for packaging, will for the first time create a complete supply chain in the US, from the start of wafer production until there is a finished packaged AI chip. This is a structural change, not just an expansion of scale.

And the reality is that this investment didn't come about in a vacuum. It comes with the US-Taiwan trade agreement, in which the US cuts tariffs on Taiwanese goods to 15%, in exchange for Taiwan committing $250 billion to invest in the US. Wei's announcement also comes just about two weeks after President Donald Trump said that TSMC was doubling the size of its operations in Arizona. What I noticed is the way TSMC expressed its commitment: tied to market demand instead of a fixed schedule, meaning that the company both satisfied political requirements and did not have to commit capital to a potentially wrong timeline. To allay concerns that pouring money into the US will empty production capacity at home, Wei also added that TSMC is planning to build 13 advanced factories and packaging facilities in Taiwan over the next few years.

Business record, but still cautious when looking ahead

The investment announcement that TSMC made came with a fifth consecutive record business quarter: net profit reached NT$706.56 billion, or about $22.35 billion, up 77.4% over the same period last year; revenue increased 36% to NT$1.27 trillion; Gross profit margin reached a record level of 67.7%. High performance computing (HPC - High Performance Computing, a group of chips serving servers and AI) accounts for 66% of revenue by platform. Furthermore, TSMC also raised its third quarter revenue forecast to about $44.6-45.8 billion, and raised its 2026 investment budget to $60-64 billion, higher than the $52-56 billion previously set. The above numbers are nice, but Wei remains cautious when talking about the longer term: he believes demand will remain strong until 2029 or 2030, but frankly admits "whether there will be a slowdown period in between, I'm not sure." A short statement but it shows that even the company that is benefiting the most from the AI ​​fever does not dare to commit to an unbroken straight line of growth. Pouring money is one thing, building it on schedule in Arizona is another. TSMC has faced labor shortages, water shortages, and problems related to foreign work visas at its current Arizona construction site, and those are factors that will determine how fast or slow the construction of these four new factories is, no less than market demand. Looking beyond the $100 billion figure, the real story worth watching over the next few years is whether the United States will be able to package all its advanced AI chips on its own soil, a capability that up until now has been almost entirely dependent on Taiwan. And when the US does this, what will happen if the US is no longer dependent on Taiwan?