To achieve profitability, TSMC’s new semiconductor fab in the United States needs to address a series of challenges.
After receiving chip subsidies, TSMC decided to introduce the latest technology at its Arizona facility, advancing President Joe Biden’s plan to strengthen the supply chain security of technology in the U.S. However, advanced chips that support AI tasks may still need to be produced in Asia, as the U.S. is missing several pieces in its AI chip puzzle.
According to the Financial Times, sources indicate that AMD plans to be among the first clients of TSMC’s Arizona plant, outsourcing the production of high-end GPUs and CPUs there. However, allowing customers to choose which factory to produce their chips contradicts TSMC’s existing strategy, reducing the flexibility in capacity allocation and thus compressing gross margins. Reportedly, customers wanting to use a specific facility must sign a separate contract with TSMC, which might involve higher prices or upfront deposits.
Moreover, although TSMC is willing to increase investment in the U.S. facility, the production timeline of local processes lags behind that of Taiwan. Analyst Myron Xie from SemiAnalysis noted, “We believe Nvidia wants to use the 2nm process technology by 2026, but TSMC will not introduce 2nm at its second Arizona site until 2028, which does not align with Nvidia’s timeline.”
However, accelerating the timeline would undermine TSMC’s critical advantage of achieving higher yields with new processes compared to its competitors, with Taiwanese R&D engineers playing a key role in the initial stages of the process. A person familiar with TSMC’s considerations stated, “We do not wish to delay the introduction of advanced processes to the U.S., but as the company expands new technologies, global R&D centers need nearby support, meaning we must first expand production in Taiwan.”
Additionally, the U.S. also needs to implement high-bandwidth memory (HBM) into advanced packaging technologies for GPU modules. Currently, TSMC has no plans to establish advanced packaging facilities in the U.S., partly because the production volume in Arizona is too low. Although Amkor Technology, a major U.S. IC packaging and testing firm, launched its CoWoS advanced packaging facility a year after TSMC’s similar facilities came online, hoping to bridge the gap, Amkor is unable to produce a key packaging component that connects logic ICs with memory.
It is noteworthy that The Wall Street Journal previously reported that Samsung Electronics plans to announce an additional investment of $20 billion in Texas next week, including the construction of an advanced packaging facility using technology similar to TSMC’s for packaging Nvidia’s AI chips.
Should TSMC shift more core assets?
Chip manufacturing is not only a pinnacle of technology and industry but also a game of costs.
This is why, although integrated circuits were invented in the U.S., their production quickly shifted to East Asia over the decades, resulting in 80% of the world’s integrated circuits being produced in Asia.
In addition to fabs, TSMC has established R&D centers in Taiwan to develop future-oriented 2nm and 1.4nm chip production technologies. TSMC’s dominant market position also influences the flow of talent.
Due to the substantial investment in fab construction, TSMC has even spurred growth in Taiwan’s construction industry.
A robust semiconductor industry and educational system have created a feedback loop, with graduates streaming into chip manufacturing companies like TSMC and UMC, continuously supplying ammunition to the industry.
This is why SemiAnalysis believes that as chip packaging technology advances, TSMC clients like Apple and Nvidia will find it increasingly difficult to decouple from Asia. TSMC always develops the latest manufacturing and packaging processes in Taiwan, where costs are lower and talent is more readily available.
This means that in the process of industrial transfer, not only factories and production lines are moved, but also the accompanying systems of technological innovation and talent development. In the context of “manufacturing reshoring,” factories can be relocated, but the talent system cannot be simply copied and pasted.
From 1979 to 2011, the U.S. lost 40% of its manufacturing jobs, likely along with the disappearance of complementary suppliers, university programs, and academic research related to the industry.
Talent will always flow to the industry’s heights, and today, American graduates in Seattle are more likely to work for Amazon than to fasten screws at a Boeing factory 23 kilometers away.
“Semiconductor manufacturing” is even an unfamiliar career choice for many Americans. Kweilin Waller, a human resources officer in Phoenix, Arizona, mentioned that job applicants often find the semiconductor industry puzzling and unfamiliar.
Thus, the U.S. attempt to reshape its chip production capabilities through subsidies and new fabs will ultimately face a talent gap.
Among the most valuable companies in the U.S., only Apple barely qualifies as a manufacturing company, yet it does not own a factory. In 2012, when Obama asked Steve Jobs what it would take to bring iPhone production back to the U.S., Jobs promptly replied, “Those jobs are not coming back.”
Disclaimer: This article is created by the original author. The content of the article represents their personal opinions. Our reposting is for sharing and discussion purposes only and does not imply our endorsement or agreement. If you have any objections, please contact us through the provided channels.
