TSMC expands investments in the U.S. to $165 billion with new fabs and R&D center: A closer look

On Monday, TSMC announced plans to invest $100 billion to expand its manufacturing capacity in the United States. The funds will go toward three additional fabs, two advanced packaging plants, and a major R&D facility. Tom's Hardware spoke with the company to learn more about its plans.

The extra $100 billion investment adds to TSMC’s $65 billion commitment for its Fab 21 site near Phoenix, Arizona. This makes TSMC, the world’s largest contract chip manufacturer, one of the biggest foreign investors in the U.S. to date.

Given the titanic proportions of TSMC's $165 billion total investment, a deeper understanding of where and how the funds are being used across TSMC's various U.S. initiatives is warranted. Let's take a look at how this fits in with TSMC's overall strategy.

Latest Videos From

Watch full video here:

New Fabs, Packaging Facilities, and an R&D Center

Although TSMC said it would invest an additional $100 billion in its U.S. operations, the company has not disclosed specific details regarding the timing, locations, or technologies for its expanded U.S. investment. But it appears to have enough space to build new facilities at its Fab 21 site.

"We have not announced details on timing, locations, or specific technologies for our intended new investments yet," Nina Kao, head of public relations at TSMC, told Tom's Hardware. "We are committed to moving as fast as possible to meet our customers’ needs, and we expect to share more information as our plans are finalized."

TSMC's Fab 21 campus near Phoenix, Arizona, is around 1,100 acres in size (4.5 km^2), which is more than twice the size of Monaco or roughly equivalent to 630 football fields placed side by side. The company originally intended to build six fab modules (or phases) there, making the facility one of the largest semiconductor production sites in the world.

When TSMC finalized its CHIPS deal with the U.S. government last year, it outlined plans to build three Fab 21 phases by 2030. Phase one includes equipment to fabricate N5 and N4 process nodes, which are already in mass production. Phase two is set to become operational in 2028 with N3 capabilities; Phase three will introduce N2 and A16 process nodes by 2030.

The new announcement adds three more Fab 21 phases, two advanced packaging facilities, and an R&D center. TSMC hopes to build everything at its Fab 21 site, making it one of its major production hubs.

"We originally selected the site in Phoenix, Arizona, and purchased a piece of land that is bigger than 1,100 acres because we would like to accommodate several operating fabs and support future expansion to scale for the economics," Kao told us. "We will work closely with the City, the State, federal government and our local infrastructure and education partners to ensure that our expansion plans can be supported."

While TSMC has not formally announced any timing for its new facilities, its latest estimate of 40,000 construction jobs over four years suggests a significant increase over the earlier projection of 20,000 unique jobs by the end of the decade.

TSMC indicates that some projects will likely proceed in parallel, leading to higher labor demand, but has not specified whether this means doubling production capacity for N3 and/or N2/A16 technology in the U.S.

"While we have not announced details of timing yet, we would expect some of these projects to happen in parallel, which would increase the construction demands," said Kao.

Building fabs in parallel could impact their timing. For example, if TSMC secures enough tools to equip its N3 (3nm-class) and N2/A16 (2nm-class, 1.6nm-class) fabs from ASML, Applied Materials, KLA, and Lam Research earlier than planned, and hires (or relocates from Taiwan) enough qualified personnel to install them, it may be possible to accelerate the completion of Fab 21 phase two and/or Fab 21 phase three.

Massive investments

Given the gargantuan investment, in addition to the projected doubling of workers at the Fab 21 site in the next four years, it's safe to say that TSMC is accelerating the number of workers at the Arizona campus. The increased construction effort is unlikely to mean a doubled investment from $65 billion to $130 billion over the next four years, however.

This is because the company might want to scale its other efforts worldwide, namely in Taiwan, Japan, and Germany. TSMC is expected to spend $38 billion to $42 billion in 2025, and while it will dramatically increase its investments in the U.S., it remains to be seen whether the company will allocate more to its Arizona-based Fab 21 than to its global facilities in Germany, Japan, and Taiwan combined.

TSMC is currently building a fab in Germany (albeit in its ESMC partnership with Bosch, Infineon, and NXP), and is about to start construction of its second fab in Japan (in its JASM collaboration with Sony and Toyota).

As for the company’s operations in Taiwan, TSMC is preparing to ramp up its N2-capable Fab 20 in the coming months. The facility is located adjacent to the R1 R&D center—which developed TSMC’s N2 nodes and their successor—near Baoshan in Hsinchu County.

TSMC’s second N2-capable fab is in the Kaohsiung Science Park, a part of the Southern Taiwan Science Park near Kaohsiung. Production at this site is expected to start around 2026. TSMC is also building two advanced packaging facilities in Taiwan. Additionally, there are rumors that TSMC is planning a 1nm-capable Fab 25—for post-N2/A16 nodes—in the Southern Taiwan Science Park near Tainan.

TSMC's leading-edge in the U.S.? 

Building out a gigafab-class site in the U.S. along with two advanced packaging facilities nearby will certainly affect where TSMC builds its chips for American companies, namely Apple, AMD, Broadcom, Nvidia, and Qualcomm. However, the question is whether is whether TSMC would produce chips using its latest technologies in the U.S. or not.

Earlier this year, Taiwan's government changed its export rules, and now TSMC is allowed to export its leading-edge production nodes to facilities overseas. Formally, this opens doors for TSMC to export its most advanced process technologies to the U.S. and perhaps Japan.

However, as TSMC develops its fabrication technologies in Taiwan, it is optimal to perform yield ramp (improvements) and production ramp-up in Taiwan, where the developers of these manufacturing processes are located. It should be noted that engineers continue to refine production nodes and reduce defect density for quarters after mass production is initiated.

Last year, TSMC gave a peak at its Global GigaFab program that enables the company to port a process technology from one fab to another quickly while maintaining achieved continuous process improvements (CPI) to increase yields, as well as statistical process control (SPC) to reduce performance variations. So, the company seems able to tune its nodes in Taiwan and simultaneously get the same results in the U.S. However, it does not mean TSMC can port a new node to a new fab overnight.

Process porting involves porting a fab layout, its settings, and raw materials. The new fab must have (or install) equipment that meets the same specs (i.e., be tuned appropriately) as the original tools. Also, the new fab must use the same raw materials as the original fab. Even slight differences in deposition methods, etch profiles, or temperature uniformities can force re-qualification of multiple steps, which renders achievements of the original fab obsolete and requires another lengthy yield ramp process.

Overall, porting a brand-new technology process from one fab to another may take 12-18 months, provided they are configured similarly. With the Global GigaFab program, TSMC can probably cut down that time by several quarters. The good news is that once a fab is configured for a process design kit (e.g., N5, which includes N5P, N4, N4P, and N4X), making chips on other nodes from the same PDK is relatively easy.

While several months delay means that Apple will continue to make its iPhone processors in Taiwan, its subsequent processors based on the same PDK and aimed at more expensive devices could be made in the U.S. By contrast, AMD and Nvidia use proven fabrication technologies, so a significant part of their products could be made in the U.S. in the second half of this decade.

It should be noted that TSMC has historically ramped up its new process technologies in Taiwan because that is where its R&D facilities are. But now that TSMC is adding an American R&D center, it might use the new center to develop certain production nodes in the U.S., then ramp them in Arizona, and volume produce on them in Arizona.

U.S. manufacturing comes at a premium

But this manufacturing will come at a cost. According to unofficial information, chips built on TSMC's N4 and N5 process nodes in the United States could command a 20% – 30% premium over their Taiwan-made counterparts. Meanwhile, more mature processes produced at the Kumamoto facility in Japan — such as N28/N22 and N16/N12 — are expected to cost 10% – 15% more when compared to similar chips manufactured in Taiwan.

That 30% premium is larger than a potential 25% tariff that the Trump administration threatened to impose on Taiwan-made chips. Of course, the situation would be different if the tariffs were higher. However, imposing a 50% import tax on Taiwan-made chips and making those products 50% more expensive for American customers is unlikely.

Nonetheless, a 20% to 30% premium over Taiwan-made chips means that not all American companies will be interested in making chips at Fab 21 unless absolutely necessary, though the company has indicated that its production capacity is sold out in the US until 2027. During its recent investment announcement, the company also thanked Apple, Nvidia, AMD, Broadcom, and Qualcomm for supporting its new US operations, implying that those companies are all customers of the Arizona fabs (specific customers' production locations are closely guarded secrets that TSMC does not reveal directly).

However, considering the premium, TSMC is less incentivized to port older fabrication technologies to its U.S. facilities. It remains to be seen how Apple, AMD, Broadcom, Nvidia, and Qualcomm will compete against rivals that will not use TSMC's services in the U.S. However, managers at those companies will likely have ideas on how to manage those additional costs. It's also possible that the Trump administration will further loosen some of the existing regulations that lead to increased costs.

Analysis

TSMC is expanding its U.S. manufacturing investment by an additional $100 billion, bringing its total commitment to $165 billion. This expansion includes three new fabs, two advanced packaging facilities, and an R&D center, primarily at its Fab 21 site in Phoenix, Arizona.

The company can potentially run projects in parallel, which could speed up the deployment of TSMC's fab in the U.S. However, despite Taiwan now allowing the export of leading-edge nodes, TSMC's preference for ramping up new technologies domestically means that U.S. fabs may experience several months of delays in adopting the latest process nodes until the appropriate capabilities are firmly established in the US. That could be the impetus for the company's new R&D facility at the Arizona campus, but only time will tell if that is the purpose; options abound.

While the investment significantly boosts U.S. semiconductor capacity, manufacturing in the U.S. will come at a premium, with chips produced at Fab 21 expected to cost 20% – 30% more than those made in Taiwan. This higher cost could limit how many companies manufacture their chips in the U.S. unless absolutely necessary.