Kevin Xu is a tech executive, former Obama White House staffer and author of Interconnected, a bilingual newsletter on the intersections of technology, business and geopolitics.
When the Taiwan Semiconductor Manufacturing Company’s Arizona fabrication facility held its “tool-in” ceremony last December, Morris Chang, who founded the company in 1987, called this event the “end of the beginning” — as in the end of the beginning of TSMC’s $40 billion investment in Arizona, but also the “end” of the beginning of deglobalization. Even though the tide of deglobalization, friendshoring and onshoring has been gaining strength for many years, TSMC’s foray into the U.S. is a landmark moment of this important shift.
As Chang proclaimed: “Globalization is almost dead. Free trade is almost dead. And a lot of people still wish they would come back, but I really don’t think they will be back for a while.”
The semiconductor industry was globalizing long before “globalization” became the norm, as Chris Miller chronicled in his book, “Chip War.” More than perhaps any other company, TSMC epitomizes all the forces of globalization — from free trade and hyper-specialization to international supply chains anchored on a presumed sense of geopolitical stability. When globalization was all well and good, these forces interacted and interconnected without much friction or impediments, propelling companies like TSMC to incredible success.
Now that TSMC is at the center of the reverse trend, its endeavors in Arizona also serve as a case study of how friendshoring really works and how much deglobalization really costs. The chip-making juggernaut revealed some of these details in its Q4 2022 earnings report.
The headline takeaway: TSMC’s U.S. fabs could cost four or five times more than its Taiwan-based counterparts.
Let’s peel the onion back further to understand where this seemingly astronomical new cost comes from and why and who will bear it. From there we can extrapolate TSMC’s experience to deepen our understanding of other strategic industries that will be impacted by deglobalization and friendshoring in similar ways.
From What And Why?
During the earnings call, Wendell Huang, TSMC’s CFO, was asked repeatedly about the cost difference between Arizona and Taiwan. His response was:
We’re not able to share with you a specific cost gap number between Taiwan and U.S., but we can share with you that the major reason for the cost gap is the construction cost of building and facilities, which can be four to five times greater for [a] U.S. fab versus a fab in Taiwan. The high cost of construction includes labor cost, cost of permits, cost of occupational safety and health regulations, inflationary costs in recent years and people and learning curve costs.
Huang also specified that this increased cost will last “a few years”; it is by no means a one-time cost or temporary problem. Unpacking Huang’s statement further, this cost largely falls under three categories: people, regulatory and knowledge. Each has its own reasons and nuances.
This category is perhaps the most straightforward to understand, though it has some uniquely “American” elements too. Building and operating fabs in Arizona will subject TSMC to higher labor costs in the form of salary, benefits and labor relations. Applying the median software engineer salary as a heuristic, the U.S. salary is roughly three times more than Taiwan’s. Employer-assumed benefits like healthcare, insurance and retirement contributions usually add another 25-30% to each employee. Healthcare, in particular, may present a jarring reality to TSMC — Taiwan has arguably one of the most cost-effective healthcare systems in the world, whereas the U.S. most certainly does not.
Regarding labor relations, the semiconductor manufacturing industry has historically been anti-union, not unlike most manufacturing sectors. This is one of the primary reasons why chip making shifted to Asia: Taiwan, South Korea, Malaysia and China. Thus, friendshoring back to the U.S. revives the specter of either possible unionization down the road or the need to fend off collective bargaining attempts, both of which incur additional costs in their own way.
The relative strengths and weaknesses of labor relations depend on the state. Some are legally and culturally more union friendly, others less so. Arizona is ranked 40th in its percentage of union workers — a not so union-friendly state. This likely played a role in TSMC selecting Arizona as its “American home.”
On the other hand, Ohio, the location of Intel’s planned $20 billion fab, is ranked 13th — a more pro-union state. Interestingly, Intel already has a sizable footprint in Arizona but decided to expand in Ohio instead. Keep in mind that most of this semiconductor-focused friendshoring is happening under an unabashedly pro-union U.S. president. How many “good-paying union jobs” a fab creates may be a factor in how much of the $39 billion manufacturing incentives from the CHIPS and Science Act Intel or TSMC or another company would receive to subsidize its investment.
A state’s union-friendliness, when contextualized in domestic political conditions, along with higher wages and benefits, presents a uniquely American set of costly tradeoffs for foreign semiconductor fab builders.
What’s more complex are the myriad legal hurdles and the multitudes of government agencies involved in “regulating” a massive greenfield project like TSMC’s new fabs.
How massive? TSMC acquired more than 1,000 acres of land north of Phoenix to house its ambitions in America. This land is large enough for six fabs (though only two are being planned right now). This land is also designed to bring TSMC’s Grand Alliance of partners and suppliers physically closer, while allowing for enough space to build any new roads, housing and other construction that TSMC deems necessary. The goal is to recreate, as much as possible, the physically tight and operationally efficient ecosystem that made Taiwan, a small island, uniquely suited for high-quality, high-capacity chipmaking.
Chang explained what made Taiwan uniquely suited to semiconductor manufacturing in a speech in 2021. Among its many advantages, he described in great detail how Taiwan’s high-speed rail system enabled the nimble deployment of thousands of workers between Hsinchu, Tainan and Taichung — TSMC’s three manufacturing hubs. High-speed rail can carry a large number of workers from one location to another in a day or less, which gives TSMC the ability to maximize worker productivity and output without needing to worry about losing talent or employee attrition, because families don’t need to relocate.
Recreating that environment is a tall order for Arizona. California’s high-speed rail debacle is a cautionary tale: Costs ballooned and progress slowed largely due to horse-trading between state, local and county officials, all of whom saw the project as an opportunity for political enrichment.
In a federalist system, where governance is separated into multiple layers that each have a certain level of autonomy, this division of labor produces artificial inefficiencies, red tape and many opportunities for “regulatory capture” — where regulators behave to advance the commercial or ideological goal of a narrow special interest, not the broader public interest. Such costly bureaucratic processes are not nearly as big of a problem in places where large development projects are centrally executed, like Taiwan and South Korea — both of which, not coincidentally, are chip-making powerhouses.
Huang’s answer alluded to the cost of “permits” and “occupational safety and health regulations” as just some examples of the compliance hurdles that TSMC must jump through, which a parade of officials in the state, county and city governments will all play a role in.
To be clear, I am not arguing against applying regulatory tools to ensure healthier working conditions and safer buildings. Incurring some regulatory cost is a worthwhile tradeoff for these protections and assurances. There are also governance benefits to a federalist system, where lower-level officials are empowered to make decisions most sensible to their local conditions.
However, the cost of navigating a complex bureaucracy is often invisible and hard to quantify but can become a major impediment to progress. A proactive stance in smoothing these hurdles from Washington would no doubt accelerate TSMC’s momentum and more quickly deliver on the promise of friendshoring.
The knowledge cost is arguably the most insidious and difficult to fix of the three categories. While people cost can be ameliorated with direct subsidies or tax credits and regulatory cost can be reduced with good lobbying or top-down government coordination, there is no policy shortcut to reversing a “brain drain” many decades in the making.
Huang diplomatically framed this problem as “learning curve costs.” But Chang was more blunt when addressing this issue in his speech during the “tool-in” ceremony: “We hired almost 600 engineers here a year and a half ago, we sent them to Taiwan, and they were under training in Taiwan for one year to a year and a half. In the meantime, about the same number of Taiwan engineers underwent training in Taiwan also. So before we see a single wafer, we have about more than a thousand people being trained. This, I think, is a very good sign that we are prepared.”
Putting it differently, TSMC has been having such a hard time finding enough quality recruits to hire that it is resorting to bringing on less-qualified workers and dispatching them for lengthy training in Taiwan. And it is hiring just as many engineers in Taiwan to go through the same process and be dispatched to Arizona to further close the knowledge gap.
The irony is that TSMC is still probably paying more for the less-qualified Americans than the Taiwanese workers, due to the salary differential between the U.S. and Taiwan alone. While higher wages and more benefits are often justified when there is market demand for certain skills, TSMC faces paying more for less-skilled workers and there are still not enough of them in America.
The only reason why Chang believes incurring this knowledge cost is “a very good sign” now is because it at least rectifies his past mistake when building TSMC’s first American fab 25 years ago. Back then, the same knowledge gap already existed. Twenty-five years later, it remains and might be worse. But at least TSMC is preemptively addressing the problem.
This challenge is not unique in the semiconductor industry — it’s systemic in U.S. manufacturing. According to a study by Deloitte and the Manufacturing Institute, the overall manufacturing skills gap in the U.S. could result in 2.1 million unfilled jobs by 2030. The surveyed executives are having trouble filling both unskilled entry-level openings and middle-skill roles across many functions.
At least there is funding in the CHIPS and Science Act — $13.2 billion — earmarked for R&D and workforce development. A nationwide skills training program to fill knowledge gaps, starting with areas most relevant to semiconductor manufacturing and expanding from there, is urgently needed.
Who Will Bear The Costs?
Many of the elements that cause the dramatic increase in the costs of doing business in the U.S. these days are somewhat predictable. What’s less predictable is who will bear them. Let’s first look at TSMC as a stand-alone case, then extrapolate to other industries that may make good friendshoring candidates.
Being the clear leader in a highly strategic industry, TSMC has strong leverage and pricing power. And it intends to flex its muscles. In the earnings call, both C.C. Wei, the TSMC CEO, and Huang, the CFO, telegraphed the company’s intention to increase prices to make up for its costly new investments in the U.S., as well as in Japan. TSMC also intends to keep hitting its long-term gross margin target of 53%, even with this price hike. Framing it as “the value of geographic flexibilities,” TSMC believes its largest American customers like Apple, Nvidia and AMD can and should bear the cost increase, if only because they (or their home government) demanded these flexibilities.
Whether Apple, Nvidia or AMD pass this on to customers in the form of pricier iPhones, gaming consoles or PCs is still to be determined. But that’s of little concern to TSMC. In many ways, TSMC’s willingness to bear the cost increase is indicative of its pricing power: The company is confident it can make its money back.
However, most industries are not as strategically important as semiconductors. So what TSMC is doing should not be over-generalized.
Many sectors may be “nice to have” candidates for friendshoring or onshoring, but deglobalization does not completely overthrow all the basic logic of unit economics, especially for sectors where the competition is high but profit margins and required workforce skills relatively low.
The story of Mastercraft Furniture, a former Oregon-based furniture maker, is one such example. Years of efforts competing with Asian companies ended in Mastercraft’s bankruptcy. When foreign furniture makers need a solution to circumvent geopolitical tension, “nearshoring” to Mexico under the new North American trade pact is more than sufficient and has better economics. Other industries like automobile wheels, construction equipment components and small electronic parts have similar dynamics and look more like furniture-making than chip-making.
The one industry whose strategic importance may resemble that of semiconductors is lithium-ion batteries — a product also dominated by Chinese (CATL, BYD), South Korean (LG, Samsung) and to a lesser extent Japanese companies. Tesla, the lone American company on the leaderboard, mostly serves its own fleet of EVs and is not a supplier for other carmakers.
Some serious friendshoring is already underway. In partnership with Honda, LG is building a $4.4 billion battery plant in the U.S. SK Group, another large South Korean battery producer, is pouring $7 billion into building gigafactories in Tennessee and Kentucky — a joint venture with Ford. Going against the tide of worsening tension between the U.S. and China, Ford is also building a new $3.5 billion battery plant in Michigan with CATL’s technology. Meanwhile, GM is partnering with Lithium Americas (an interesting friendshoring candidate that has strong ties to Argentina and whose stock is trading on the Toronto Stock Exchange) to mine lithium in Nevada for batteries.
With the Inflation Reduction Act’s subsidies for EV purchases and an industry-wide price reduction, consumers have not had to pay more for the friendshoring of lithium-ion battery production just yet. Of course, a federal subsidy is simply a reprioritization of tax dollars already levied from taxpayers, who are always the ultimate bearers of these costs. The only question is whether reprioritizing this money to building strategic redundancy instead of funding new innovation, all in the name of national security and competitiveness, is worth the price.
How To Be Good Friends
Deglobalization and friendshoring are surely here to stay for the near future, especially in strategic industries. However, there is no official definition of what “strategic” means. It all depends on shifting geopolitical relationships and domestic political dynamics. In the context of rivalry between the U.S. and China, what is “strategic” appears to be whatever China wants but can’t have or whatever the U.S. wants that China has a lot of.
There is also no official definition of what a “friend” is. While grandiose alliance frameworks like the Quad and the Chip 4 initiative delineate who is friends with whom, being a “good friend” in the age of deglobalization requires more than regular summits and photo-ops.
TSMC’s adventure in Arizona provides a roadmap of how the U.S. can be a “good friend”: invest resources to closing the knowledge gap, not playing favors between pro- and anti-union states and minimizing red tape through tighter top-down alignment along all levels of government in order to speed up TSMC’s progress. With an inherently competitive federalist system, the U.S. has its work cut out.
If the U.S. is not a “good friend,” it could get out-competed by others. TSMC is simultaneously friendshoring in Japan, who once dominated DRAM semiconductor manufacturing in the 1980s at the U.S.’s expense and has a better track record of top-down industrial policy alignment. Friends can become “frenemies” quickly in the age of deglobalization.
TSMC has shown us how much deglobalization truly costs. The next challenge is making geopolitical “friendship” economically more sensible too.