Semiconductors are silicon oil. Chips are the lifeblood of the largest global corporations and absolutely critical to cutting-edge AI infrastructure, national security, and military capabilities.

And one small Maryland-sized island southeast of China dominates their production. Today, Taiwan accounts for 65% of the world’s semiconductors, and 90% of its most advanced chips.¹

Recognizing this strategic vulnerability, the U.S. passed the bipartisan CHIPS Act in 2022 — a $280 billion bet to return chip manufacturing to American soil.² The Act allocated $52.7 billion to support the semiconductor industry — including $39 billion for manufacturing incentives, $11 billion for R&D, and another $500 million for multilateral cooperation.³

The legislation has incentivized international chip giants to make some of the largest domestic investments in American history.⁴ Since CHIPS passed, Intel, Micron, and the Taiwan Semiconductor Manufacturing Company (TSMC) have pledged to invest tens of billions into factories in Ohio, New York, and Arizona.⁵ The authors contend that is the problem.

This summer the authors spent time in Hsinchu, Taiwan, home to companies that today rule an industry the United States once called its own, seeking to understand what makes their chip ecosystem so effective. The answer: semiconductor success is not found solely in technology. It’s in the space between things — or rather, the lack of it.

Within a radius the size of an American suburb, Hsinchu houses many of Taiwan’s national research laboratories, elite universities, and sprawling semiconductor fabrication facilities — fabs. Proximity allows ideas and internships to flow freely; this is routine. If the United States is trying to replicate what makes Taiwan tick, a couple independent chip factories is insufficient — the country needs to build dense, semiconductor cities.

The question becomes: what makes a world-class “science park” and how can America create one?

Vertical Education Pipeline

Hsinchu Science Park is a short walk from National Yang Ming Chiao Tung University (NYCU). There, the authors met with one of Taiwan’s leading semiconductor professors who requested to remain anonymous for security reasons, to understand TSMC’s hold on the regional ecosystem. “Much of our research is guided by TSMC”, he commented. “They send us projects they want us to explore and fund them.” With fabs focused on production, it is little surprise TSMC outsources intellectual work to the country’s top minds — minds that are by no accident across the street.

Similarly, TSMC funds government labs to undertake projects in parallel, further utilizing the local talent hub. “You’ll often have TSMC researchers visiting the schools and labs directly and overseeing those projects. Many of the students I mentored over the years have ended up at TSMC, many with decade-long tenures.” NYCU’s students live across the street from these semiconductor facilities — some of the most advanced in the world — where they’re encouraged to work. Ideas move quickly from research labs to production fabs.

In the cafeterias where students break bread with TSMC engineers and attend NVIDIA recruiting events, future chip makers can see their destiny from their dormitories. But Taiwan’s pipeline actually starts long before then.

“TSMC funds K-12 education programs,” the professor explained. “They’ll bring chips into a classroom of eight-year-olds and get them inspired by the technology’s seemingly magical capabilities. It builds up a culture of excitement from a very, very young age, where manufacturing is perceived as important and prestigious.”

Many of these students pursue electrical engineering in their undergraduate and postgraduate educations, a path that equips them well for roles at TSMC. Their learning is often supplemented by vocational chip-manufacturing programs, also funded by the tech giant.

Technical vocational programs are not a rarity among large manufacturing organizations outside Taiwan. Both Toyota and Dyson offer undergraduate programs that prepare students for guaranteed careers at their factories.⁶ But these programs often stipulate that students must work for the specific companies that fund them.

TSMC’s programs, on the other hand, prepare students for broader higher education at nationally accredited universities. While they may lose some talent they’ve invested in to competitors, those that do ultimately end up on their teams are exceptionally technical engineers with broad understanding of design processes. The universities they invest in located in or near Hsinchu, such as NYCU, National Tsing Hua University, and NTU, are some of Taiwan’s best.

By contrast, Arizona — where TSMC is putting down roots for its flagship American fab — ranks 49th in the country in K-12 per-pupil funding.⁷ While making direct comparisons between the American and Taiwanese education systems is difficult, the majority of employees at American fabs are not yet benefiting from a similar local talent pipeline.

The issue transcends any singular project or locale. While perhaps not the perfect location to center chip production, Arizona could become a thriving semiconductor hub over the coming decade. America’s scattered approach is the larger concern; by spreading investments across multiple regions of the country, the United States risks diluting its reshoring efforts.

Later, as on the grounds of Hsinchu’s semiconductor factories, the authors asked those wearing the dark maroon TSMC badge where they had studied.

“Oh, right here in Hsinchu”.

“NYCU”.

“A couple kilometers down that street”.

Karman Lucero, a Fellow at the Paul Tsai China Center at Yale Law School with whom the authors spoke, recounted an episode relating to a critical part breakdown at a fab in Taiwan. “This could shut down production for hours, and folks are expected to be on call to handle incidents like this at all times,” he said. When it happened, the engineer on call simply called a friend, a former classmate who now works at a rival company, for help. The part was rapidly replaced.

Education, research, work — all that talent aggregated across several blocks.

Bureaucratic, Geographic Advantage

Proximity also reduces red tape. In Kaohsiung, a southern province of Taiwan, the city government expedited over two years of regulatory and construction pre-work to clear a site for a new advanced fab before confirming it would be built. Regulations were quickly cleared, with the mayor declaring that “opportunities are reserved for those who are prepared.”⁸

Opportunities also come to those nearby. Over a lunch, Kuan-cheng Huang, a logistics professor and government transportation advisor, pulled up a map of the global semiconductor supply chain. The same pattern is repeated on a larger scale: the efficiency between buildings in Hsinchu is replicated between cities across Asia.⁹

Chips originate as equipment, photomasks, and chemical imports from Taiwan’s neighbors, most notably Japan and Korea. From there, the components are processed by Taiwan’s foundries. Before an original equipment manufacturer (OEM) like Apple can do anything with these chips, they are sent through design manufacturers and electronic management services. These entities are responsible for circuit board assembly and design engineering, as well as testing chips through advanced, IP-protected quality control. They are completely irreplaceable parts of the chip manufacturing process.

Out of the largest of these companies, Foxconn and Wistron are based in Taiwan; BYD and USI are headquartered in China; Flex is Singaporean; and Jabil, despite being American, has a significant number of employees operating in Asia.¹⁰ Geographically, Taiwan lies in the epicenter of this intricate network.

Advanced packaging — the intricate integration of semiconductor dies with electrical connections — is a high-value bottleneck of the manufacturing process, particularly for scaling GPU production. Chips produced in Arizona will be shipped back to Taiwan primarily for this purpose.¹¹ While “applaud[ing] Amkor” for its recent $2 billion packaging and testing investment in Peoria, Arizona, as of January 2024, TSMC has yet to move any of their own packaging capacity into the state.¹²

Professor Huang also emphasized how tremendously damaging even micro-errors can be to the ecosystem. “Every step needs to be perfect. When we’re scaling up to millions or even billions of transistors, a failure rate of 0.01% to 0.001% could be the difference between successful chipmaking and catastrophic failure.” That sort of precision, Huang says, comes with time.

It also comes with people. “You have engineers who’ve spent decades of their lives optimizing their individual machines for perfection…they’ve had 30, 40 years to fine tune every individual detail.”

South Korea is currently following a similar playbook of employing geographic concentration and efficiency as competitive advantages. The new Yongin semiconductor cluster, which will be 3.6 times the size of Monaco, was approved three months ahead of schedule.¹³ By positioning the cluster near Samsung’s existing facilities, power plants, emerging residential areas, and fab suppliers, South Korea aims to replicate Taiwan’s advantages of proximity.¹⁴

The Manhattan Project Model

Tim Culpan, an independent journalist who has covered Taiwan’s tech scene since 1999, believes “U.S. policymakers need to understand that complete supply chain independence is a myth. Instead, they should focus on building a small amount of redundancy while also ensuring security and stability among its overseas partners.”

In the short term, this seems like the most feasible approach. The economic and geopolitical benefits of reshoring at least part of the semiconductor industry are real. Doing so could create millions of accessible jobs and virtuous cycles of economic growth, while reducing American dependence on foreign partners. These steps would also provide greater security for the booming AI industry. The stakes are high, as publicly available estimates indicate that a one-year loss of access to TSMC fabs would cause around $1.6 trillion in economic losses for the United States, shaving a massive 8% off annual GDP.¹⁵

Globally, such a loss could precipitate a wider economic downturn. Domestically, it would cripple military readiness and compromise critical infrastructure. Department of Defense weapons systems, from F-35s to missile defense, rely on advanced semiconductors. Cutting-edge semiconductors are even more essential to emerging AI capabilities. Manufacturing redundancy is a small price to pay to hedge against the risk of losing access to this critical input.

America has proven this model before. Los Alamos — a small city in the desert with schools, homes, and hospitals — proved how concentrating scientific talent could accelerate innovation for national security. Over the last few decades, Silicon Valley has shown how geographic density creates unparalleled technological prowess. San Francisco exemplifies this today — the AI revolution is driven by an incredible vortex of talent in one city. While America’s new fabs do not need to be built in the Bay Area, there’s synergistic value in proximity, whether in Arizona, Ohio, New York, or somewhere else.

Corporate objectives and the realities of domestic politics present challenges to this approach. It is essential to align Americans’ interests nationwide with doubling down on localization. To incentivize politicians from various regions to support concentrated chip production investment in Arizona as part of a cohesive national strategy, policy proposals could reframe subsidies as security measures and distribute economic benefits across multiple states.

That is, the United States can support multiple hubs focused on different areas of the AI ecosystem while centralizing chip production itself. Taiwan’s ecosystem is condensed by geographic necessity, whereas compute clusters, data centers, and manufacturing facilities that benefit from a burgeoning domestic chip industry could be established in other parts of America. This means private investment through initiatives like Stargate and public research funding for university towns distributed nationally.¹⁶

Professor Ted Wittenstein of the Yale Schmidt Program on AI offers one historical analogy: “If you look at why the U.S. created the Strategic Petroleum Reserve, it came from the sense in the 1970’s that we were highly vulnerable to global disruptions. As we are unlikely to replace our demand from Taiwan, the question becomes what milestones we could expect for a comparable chip alternative.” While the economics are complex, a ‘Chip Reserve’ could stimulate domestic demand and be strategically placed in a region of the United States, such as the Midwest, that might otherwise gain little from increased investment in Arizona.

To be clear, these complementary elements, from compute clusters to a chip reserve, don’t require tight geographic concentration with chip production. Educational programs do.¹⁷ Components tied directly to semiconductor fabrication, like labor talent pipelines and core production facilities, must be localized for a successful ecosystem. As such, targeted public-private educational investments in electrical engineering and chip research and development should be organized around chip manufacturing hubs.

The Manhattan project worked for a reason. Today’s semiconductor challenge demands a similar strategy.

An American “Innovation Stack”

Taiwan’s advantage lies in its geographically concentrated “innovation stack”: a moat derived from ten to fifteen small advantages that exponentially compound into a system that is nearly impossible to replicate.¹⁸ This is how a tiny competitor has come to play such a vital role in a unique web of semiconductor innovation, a place where black magic happens in black boxes.

Taiwan boasts a vertical education pipeline with research partnerships, associated prestige, and financial reward; an intense, resourceful work culture; geographic agglomeration of facilities; and an incredible ecosystem of innovation. This is Hsinchu. On a macro scale, this is Taiwan, which stretches only a few hours from end to end by bullet train.

America is much larger geographically. It is a more complex ecosystem. For large, distributed industries, this proves advantageous. For critical innovation — the type fueled by informal meetings, talent pipelines, closed-door conversations, and rapid iterations — scale induces lag. When you are etching billions of transistors into a semiconductor every second of every day, lag compounds.¹⁹

The truth is, the chips didn’t “go” anywhere; America let them slip away — to a place where production is cheaper, more efficient, and more compact. If the United States aims to succeed with its grand ambitions in AI, it might need to start thinking small.

The current administration faces one of the greatest commercial manufacturing and national security challenges of our time. Its tenure will determine if America can develop a secure domestic semiconductor ecosystem that captures Taiwan’s magic or watch as chip production becomes permanently entrenched across the Pacific.

Quotes, anecdotes, and other specific details come from the authors’ personal notes and recordings from numerous conversations in Hsinchu, Taiwan and San Francisco, CA with industry experts, engineers, and academics.

The authors thank Professors Ted Wittenstein, Phil Kaplan, Hanscom Smith, Lauren Young; and Karman Lucero, Tim Culpan, Chris Miller, Brady Helwig, Teddy Tawil, Ethan Chiu, Katherine He, Sneha Revunar, Jenny Duan, and Mohit Agarwal for their editorial feedback and support.

1. “Taiwan’s dominance of the chip industry makes it more important,” The Economist, March 6, 2023. ↩︎
2. Justin Badlam, “The Chips and Science Act: Here’s What’s in It,” McKinsey & Company, October 4, 2022 ↩︎
3. “Fact Sheet: Chips and Science Act Will Lower Costs, Create Jobs, Strengthen Supply Chains, and Counter China,” The White House, February 3, 2023. ↩︎
4. Ibid. ↩︎
5. Paisley Coxsey, “How Multibillion-Dollar Plans from TSMC, Intel Compare to Other Gigantic Us Tech Projects,” Arizona Technology Council, May 3, 2024. ↩︎
6. “Technician Training,” n.d. Toyota; “Offer Information,” Dyson Institute, 2019. ↩︎
7. “Arizona Ranks 49th in K-12 Funding, Again,” Save Our Schools Arizona Network, December 19, 2023. ↩︎
8. “TSMC Drops Northern Taiwan Site for Advanced Chip Factory after Protests,” Reuters, October 17, 2023. ↩︎
9. RayMing, “Top 21 Largest EMS Companies in the World,” RAYPCB, 2023. ↩︎
10. Ibid. ↩︎
11. Wen-Yee Lee and Fanny Potkin, “Exclusive: TSMC in Talks with Nvidia for AI Chip Production in Arizona, Sources Say,” Reuters, December 5, 2024. ↩︎
12. Russ Wiles, “Amkor to Build $2 Billion Semiconductor Testing Plant in Peoria. 2,000 Jobs Promised,” The Arizona Republic, November 30, 2023. ↩︎
13. Kim Boram, “(Lead) Gov’t Designates Yongin Chip Cluster as National Industrial Complex,” Yonhap News Agency, December 26, 2024; Anton Shilov, “South Korea Greenlights the World’s Largest Semiconductor Hub – Half the Size of Beverly Hills to House Six Fabs, Three Power Plans, and Fab Suppliers,” Tom’s Hardware, December 27, 2024. ↩︎
14. “Govt. designates Yongin chip cluster as national industrial complex,” The Korea Herald, December 26 2024. ↩︎
15. Riley Walters, “Losing Taiwan’s Semiconductors Would Devastate the US Economy,” Hudson Institute, October 16, 2023. ↩︎
16. Craig S. Smith, “Stargate: America’s $500 Billion Bid to Corner Global AI Capital.” Forbes, January 24, 2025. ↩︎
17. “Teaching K-12 Teachers: Purdue Workshop Aims to Build Future Workforce for the Midwest Semiconductor Industry,” Purdue Polytechnic Institute, 2024. ↩︎
18. Jim McKelvey, The Innovation Stack: Building an Unbeatable Business One Crazy Idea at a Time (New York, N.Y.: Penguin Publishing Group, 2020). ↩︎
19. Matthew Griffin, “TSMC Says It Expects to Produce 1nm Trillion Transistor Chips by 2030,” Matthew Griffin | Keynote Speaker & Master Futurist, January 24, 2024 ↩︎