1. The Empty Cleanroom in Texas
July 2025. Taylor, Texas. On the edge of this small city 50 kilometers northeast of Austin, a massive structure stood at 92 percent completion. The Samsung Electronics Taylor Foundry Fab. Total investment: $44 billion — roughly ₩60 trillion. The single largest investment a Korean company had ever poured onto American soil.
A single line from Nikkei Asia summarized the plant's reality. "There are no customers."
When Samsung announced the construction of the Taylor Fab in November 2021, the city was close to festive. Taylor Mayor Bryce Smith declared it "the day that changes our city's history," and the Biden administration held up the investment as a visible achievement of the CHIPS Act. The Department of Commerce promised Samsung up to $6.4 billion — roughly ₩8.3 trillion — in direct subsidies. 2,000 permanent jobs were riding on the plant.
But as the building rose, the customers did not come. TSMC was monopolizing Apple, NVIDIA, and AMD — the companies that buy more chips than anyone else in the world. Samsung Foundry's yield problems were an open secret inside the industry. In Chapter 11 we saw Samsung Foundry's global market share sitting at just 8.1 percent. Against TSMC's 67.1 percent, this plant was built on an uneven playing field from the outset. Even within Samsung, self-deprecating remarks circulated: "We have to open the Taylor Fab if only to claim the CHIPS Act subsidy. If we don't, we fall even further behind our competitors."
In the second quarter of 2024, with construction at 92 percent, equipment installation began to slip. Samsung had decided to upgrade from 4-nanometer to 2-nanometer process technology — and in the interval, had failed to secure key customers. In April, the Department of Commerce and Samsung formally agreed on the CHIPS Act subsidy.
By December 2025, however, the situation had deteriorated further. Amid the uncertainty of Trump-era tariffs, Samsung's total U.S. investment plan was cut from $44 billion to $37 billion — roughly ₩48.1 trillion. The federal subsidy was also reduced 26 percent, from $6.4 billion to $4.75 billion — roughly ₩6.2 trillion. Production start was pushed back again, to 2026. In Chapter 4 we saw how TSMC, committing $165 billion — roughly ₩214.5 trillion — to Arizona, was exposed to the whims of American politics. The Samsung Taylor Fab is the Korean edition of that same structural vulnerability.
A building 92 percent complete. An empty cleanroom. This image shows the underside of the grand narrative called supply-chain realignment. A state can fund the construction of a factory. It cannot guarantee the customers who will fill that factory. The CHIPS Act laid bricks but could not generate purchase orders. In Chapter 1 we saw how, when the Hanseatic League collapsed, technological innovation bypassed existing supply chains. In supply-chain realignment, the winners are determined not by the size of the subsidy but by the trust of the market.
2. The Map of Separation
What the Samsung Taylor Fab reveals is not simply one company's failure. It is a cross-section of the new terrain created by the tectonic shift called U.S.–China decoupling.
The numbers speak to the pace of separation. China's share of total U.S. imports fell from 21 percent in 2018 to 13.4 percent in 2024, and then below 10 percent in the first half of 2025. Halved in seven years.
The U.S.–China bilateral trade share of global trade was also cut roughly in half, from 3.6 percent in 2015 to approximately 2 percent in 2025. Investment is even more dramatic — corporate investment between the two countries has already fallen below 1 percent of global investment flows, a state of effective separation.
Yet looking at absolute scale changes the picture. In 2025, total U.S.–China goods trade still exceeded $650 billion annually. What is underway is not complete severance but "managed decoupling." Since the tariff war began in 2018, the Biden administration imposed 100 percent tariffs on electric vehicles and 50 percent on semiconductors, and once additional tariffs on all Chinese goods were applied in the second Trump term, U.S. imports from China fell 25 percent year-on-year in the first half of 2025.
Beneath the surface, a more paradoxical phenomenon is at work. To understand it, one face is needed before the numbers.
In early 2022, David Huang — CEO of a Texas-based electronic components firm — heard the news of a third lockdown at his Guangzhou factory. Deliveries slipped again. A call came from a major customer: "We're putting a China+1 clause in the contract. Without a production base outside China, it will be hard to renew." Huang boarded a flight to an industrial park near Hanoi in northern Vietnam.
The move was far more painful than anticipated. Key managers in Guangzhou refused to relocate. Delivery times at the Hanoi park ran 30 percent late for the first three months — because sourcing components from China took longer than it did in China. Vietnam's local procurement rate for components was less than half of China's. Three years later, Huang's company settled into a "dual track" running both its Vietnam and Guangzhou plants. The customer's China+1 requirement was met. But the connection to China's supply chain could not be severed.
This is the structure behind U.S. imports from Vietnam surging while Vietnam's import dependence on China rose from 28 percent to 33 percent. Mexico's share of imports from China also rose, from 18 percent to 20 percent. China transformed from a finished-goods exporter into a supplier of intermediate goods — not vanishing from the supply chain but retreating one layer deeper into it. At the same time that Western companies are leaving China, Chinese companies are investing in Mexico to circumvent the USMCA. The paradox of decoupling: the harder you separate, the more new connections form.
The American response grew steadily more systematic. In October 2022, a sweeping revision of export regulations covering semiconductor manufacturing equipment. In October 2023, a ban extending to NVIDIA's performance-limited chips for China — the A800 and H800. The Biden administration introduced the "AI Diffusion Framework" in January 2025, dividing the world into three tiers. Tier 1 — 18 allied nations including the United States, Korea, Japan, Taiwan, and the United Kingdom — received unrestricted access to AI chips. Tier 3 — China, Russia, Iran, and North Korea — faced effective prohibition. The remaining roughly 150 countries were Tier 2, with limited access. In Chapter 7 we saw the United States flip the switch on ASML's exports through the Netherlands. The AI Diffusion Framework extended that logic from semiconductor equipment to AI chips as a whole.
The Trump administration, however, abolished the Tier structure two days before it took effect and shifted to a more flexible approach. What changed was not the direction of policy but its pace and form. Decoupling has become an irreversible current. The remaining question is not "will separation happen" but "at what speed and in what form."
The word friendshoring crystallizes this current — the strategy of shifting supply chains toward allies who share political values and trade agreements. Mexico became the United States' largest source of imports in 2024 — $505.9 billion in exports to the U.S., 15.6 percent of total American imports. Vietnam's electronics exports reached $126.5 billion in 2024. India's electronics production reached $125 billion in the same year.
The map of separation is being redrawn. Everywhere a new line is drawn, opportunity and risk intersect.
3. The Decision That Came Before the Subsidy
At this intersection, the trajectories of Korean companies split in two. One is the delay of the Samsung Taylor Fab; the other is the preemptive investment of SK On in Georgia.
Commerce, Georgia, spring 2022. SK Battery America's Plant 1 — the first large-scale standalone battery plant in the United States — began operations. The cells moving down the line were destined for the Ford F-150 Lightning. On the shop floor, Korean engineers were explaining electrode-coating tolerances to American line managers. The units didn't match — the Koreans worked in micrometers while the Americans asked for conversions to milli-inches. Closing that small gap took a month. A banner on the outer wall of the plant read: "2,600 JOBS, JACKSON COUNTY."
In March 2019, SK Innovation — now SK On — announced the groundbreaking of a battery plant in that county at a scale of $1.67 billion — roughly ₩2.2 trillion. At the time, the American EV market was still at an early stage. There were no subsidies. The IRA did not yet exist. The single thing SK was betting on was one judgment — that the American market would inevitably come.
In 2021, that judgment came under test. A battery patent dispute with LG Energy Solution brought the U.S. International Trade Commission (ITC) to the brink of issuing an import ban against SK's batteries. SK On's entire American business was at risk of disappearing. The Biden administration exercised its veto to block the ban. The stated reason was "national security interests." The 2,600 jobs already rooted in Georgia had become the lifeline. Georgia Governor Brian Kemp later announced personally that SK had exceeded its employment targets — two years ahead of schedule.
On August 16, 2022, the IRA was signed. SK On became one of its biggest beneficiaries. Not a company that had waited for the IRA — but a company that had already been on American soil before the IRA arrived — claimed the largest dividend. The BlueOval SK joint venture with Ford, three battery plants totaling $11.5 billion — roughly ₩14.9 trillion — in Kentucky and Tennessee, and a $9.63 billion — roughly ₩12.5 trillion — federal loan from the Department of Energy: the seed planted in Georgia had grown into a vast tree.
Yet this story, too, contains its paradox. In 2025, a slowdown in EV demand shook every plan. BlueOval SK's plant operating schedules were delayed, and discussions began over dissolving the joint venture with Ford. The Trump administration terminated the consumer EV tax credit. Right after Hyundai Motor announced $21 billion — roughly ₩27.3 trillion — in U.S. investments, the benefits were cut. Promise, then reversal.
In Chapter 11 we saw that the IRA was simultaneously opportunity and constraint for Korean battery companies. An opportunity created by policy is still validated by the market.
While Samsung invested a cumulative $22.4 billion — roughly ₩29.1 trillion — in Vietnam, making it responsible for 45 percent of global smartphone production, Vietnam's import dependence on China rose further — the David Huang structure we saw in Section 2 applies to large corporations in exactly the same way. The core components of "Made in Vietnam" remain "Made in China."
Korea's coordinates lie between these two trajectories. The delay at Samsung's Taylor Fab shows that supply-chain realignment cannot be achieved by will alone; SK On in Georgia proves that a preemptive decision creates policy dividends. Both cases point to the same lesson — in supply-chain realignment, timing, not subsidies, decides the winner.
4. The Gravity of 1.4 Billion — India as Variable
A new gravitational center is rising on the map of supply-chain realignment. India.
In October 2022, Apple quietly began producing the iPhone 14 at Foxconn's plant in Tamil Nadu, India. Tim Cook never announced a supply-chain diversification. He simply did it. Four years later, the numbers said everything. In 2025, 55 million iPhones manufactured in India — a 53 percent increase year-on-year. Twenty-five percent of Apple's total production had shifted to India. iPhone exports from India exceeded 2 trillion rupees — roughly $23 billion — making them India's single largest export product. Between March and May 2025, 97 percent of India-produced iPhones were shipped to the United States — concentrated shipments in the days just before Trump's tariffs took effect.
More notable still is the emergence of Tata Electronics. The Indian domestic company's share of iPhone production surged from 13 percent in 2024 to 37–40 percent in 2025. A homegrown Indian company had been incorporated into the global supply chain. This is not simply the relocation of an assembly plant. It is a signal that India is transitioning from passive recipient of supply chains to active participant.
India was not covered among the seven mirrors in Part II. This book's three existing archetypes — the small technology nation (Singapore, Israel), the resource major (Indonesia), and the sovereign-capital pivot (UAE) — none fit cleanly. A "small nation" of 1.44 billion is a definitional contradiction; there is no monopoly over a single resource like nickel or lithium; there is no sovereign wealth fund. India is a fourth type — a composite major combining market scale, low-cost highly-educated labor, and an IT services ecosystem. And with 70 years of accumulated non-alignment DNA, India simultaneously engages the United States (Quad), Russia (more than 40 percent of oil imports), and China (bilateral trade of $127.7 billion). Strategic autonomy that refuses incorporation into any single bloc — this is precisely what reveals India to be the largest practitioner of the strategy of the in-between.
For Korea, India operates as three vectors simultaneously.
India as market. A middle class that will reach approximately 500 million by 2030 is the most realistic large-market alternative after China. Samsung operates the world's largest smartphone factory in Noida, with annual capacity of up to 120 million units after the 2018 expansion, employing more than 70,000. Hyundai Motor executed a $3.3 billion — roughly ₩4.3 trillion — IPO on the Indian stock exchange in 2024 and declared India a long-term core base. It plans additional investment of $3.8 billion — roughly ₩4.9 trillion — in Tamil Nadu and Maharashtra to raise annual production capacity to 1.1 million units by 2028. Korea–India bilateral trade targets doubling, from $25.1 billion to $50 billion by 2030.
India as competitor. Through the PLI (Production Linked Incentive) scheme, 806 companies have been approved, and electronics production has grown 146 percent in five years. Smartphone imports fell from 210 million units to 3 million — domestic substitution has succeeded. Competition with Korea is forming in smartphones, home appliances, and electronic components. The fact that Tata has begun producing iPhones is evidence that Indian manufacturing is moving beyond simple assembly.
India as partner. As India's semiconductor ATMP (Assembly, Testing, Marking, and Packaging) capacity grows, a division of labor becomes possible that connects Korean memory chips to Indian back-end processes. Micron's $2.75 billion — roughly ₩3.6 trillion — investment in Gujarat to build an ATMP facility is the first step in that direction. Tata Electronics has partnered with Taiwan's Powerchip to build an $11 billion — roughly ₩14.3 trillion — semiconductor fab in Dholera, Gujarat — 28nm to 110nm processes, with a 2026 target for operations. On that construction site in Dholera, Taiwanese engineers who had spent five years as process engineers at TSMC's Nanjing fab before moving to Tata were teaching wafer transfer protocols to new Indian engineers. Training content that took one day to cover in Taiwan required three days in Dholera — because translating equipment manuals from traditional Chinese to English, and then into Hindi, consumed the time. In 2025, a Korea–India–U.S. trilateral 1.5-track dialogue was convened, discussing a technology talent academy and a Korean SME industrial complex.
Indonesia banned raw nickel ore exports and chose to force its way up the value chain. India's strategy is different. It does not compete on resources — it competes on scale. The gravitational pull of a consumer market of 1.4 billion draws supply chains toward it.
But that gravity carries friction. A Korean-heritage quality-control manager working at Foxconn's plant in Sriperumbudur, Tamil Nadu, commutes two hours each way to the factory every morning. In the monsoon season, the flooded roads made it three. In the reports he sent to headquarters in Seoul, the word "infrastructure" appeared before any defect-rate figure. Infrastructure shortfalls, bureaucracy, and the structural constraint that 90 percent of the labor force is informally employed. Eighty-three percent of the unemployed youth are between 15 and 29, and the female labor force participation rate stands at just 41.7 percent. A water crisis — sustaining 18 percent of the world's population on 4 percent of its freshwater — constrains the siting of water-intensive facilities like semiconductor fabs.
India's demographic dividend does not realize itself automatically. The number "median age 29" is potential, not guarantee. The gap between potential and reality is what makes India not a "certain alternative" but a "vast variable."
5. The Invisible Front — Three Dimensions of AI Sovereignty
If supply-chain realignment is the tectonic shift of the physical world, AI sovereignty is the territorial contest of the digital world. The two fronts converge into one structure.
AI sovereignty is a nation's capacity to develop and control its own AI infrastructure, data, and models without dependence on foreign technology providers. McKinsey estimates that by 2030, 30–40 percent of global AI spending — roughly $500–600 billion — will be shaped by sovereignty requirements. AI sovereignty is not a declaration. It is a market.
That market divides into three dimensions.
The foundation is compute. Training AI requires GPUs. NVIDIA dominates that market, and Korea, as a Tier 1 nation, has unrestricted access. In 2025, deployment of more than 250,000 NVIDIA GPUs across Korea was agreed upon: 50,000 for Samsung Electronics, 50,000 for SK Group, 50,000 for Hyundai Motor.
Data centers house those GPUs. The Ulsan AI Data Center being jointly built by SK Telecom and AWS — at a scale of $5 billion, roughly ₩6.8 trillion — will be the largest in Asia upon completion in 2027. In South Jeolla Province, a data center complex of 3 gigawatts at a scale of $35 billion — roughly ₩45.5 trillion — is planned. In the process of deciding that location, a director at the Ministry of Trade, Industry and Energy had to weigh three variables simultaneously — coastal access sufficient to secure cooling water, transmission distance from nuclear plants, and acceptance by local residents. One gigawatt of electricity is equivalent to one nuclear reactor. A single data center's location now simultaneously determines energy policy and national land planning. BlackRock and the Korean government have announced the development of a hyperscale AI data center hub. In Chapter 9 we saw the UAE use the StarGate partnership with American OpenAI to build a large-scale AI data center belt in Abu Dhabi — a strategy of using sovereign capital to acquire AI infrastructure. Korea is assembling the same infrastructure through a combination of private chaebol and foreign big tech, not a sovereign wealth fund — the character of the capital differs, but the objective of physically securing AI compute sovereignty is identical.
That infrastructure consumes power. In 2024, global data center electricity consumption reached 460 terawatt-hours. By 2030, it is projected to exceed 1,000 terawatt-hours. Goldman Sachs estimates that 85–90 gigawatts of new nuclear capacity will be needed by then. Korea's response is the SMR (Small Modular Reactor) partnership among X-energy, Amazon, Korea Hydro & Nuclear Power, and Doosan Enerbility. Doosan's decision to build an SMR manufacturing facility in Changwon illustrates the structure by which AI and nuclear power are connected into a single investment chain.
But there is a compute paradox. A country that consumes AI chips while simultaneously supplying the critical components of those AI chips — this is Korea's distinctive position. As we saw in Chapter 11, NVIDIA's H200 does not function without SK hynix's HBM. Korea's share of the HBM market — SK hynix plus Samsung — stands at 79 percent. Neither American AI dominance nor China's AI catch-up is possible without Korean-made HBM. Consumer and supplier at once. This dual position generates Korea's negotiating leverage.
The data flowing over that infrastructure is another front. The performance of an AI model depends on the quantity and quality of training data. An Asia head of a global cloud company, while examining the possibility of building a Seoul data center, wrote in an internal memo: "Korea can exchange data with the EU, it can upload to American cloud, and trade routes with China remain open. It is rare in Asia to find somewhere that satisfies all three conditions simultaneously." Here, regulation creates the terrain. Korea's Personal Information Protection Act (PIPA) is among the first legal frameworks in Asia to receive an adequacy decision from the EU. That decision, which took effect in December 2021, streamlines the transfer of personal data between Korea and the EU.
At the same time, Korea is differentiated from China's coercive data localization. China legally mandates domestic storage of "important data" through a three-law system — the Cybersecurity Law, the Data Security Law, and the Personal Information Protection Law. Korea and the EU in principle permit cross-border transfers while requiring safety measures. This difference is the structural foundation that makes Korea a candidate for a data hub.
Korea is one of the few major economies simultaneously participating in IPEF (U.S.-led) and RCEP (which includes China). Compatible with the EU while minimizing conflict with the American Cloud Act, and connected to China through RCEP. In Chapter 3 we saw Singapore build indispensability through regulatory Goldilocks — neither too loose nor too strict. Korea's data governance operates on the same logic.
The weakest link lies in the model dimension. Korea's position is most vulnerable here. Naver's HyperCLOVA X was trained on 6,500 times more Korean-language data than GPT-4. It has secured 60,000 corporate clients, and in 2025 launched HyperCLOVA X Think, a model specialized for reasoning. Samsung Gauss handles translation, summarization, and image generation on Galaxy smartphones as on-device AI. LG's Exaone, SKT's A.X, Kakao's Kanana — Korean companies are building their own models.
Yet the gap from OpenAI's GPT, Anthropic's Claude, and Google's Gemini is significant in practice. In English performance, reasoning capability, and scale of training data, Korean models do not reach the global frontier. Developing a frontier model requires investments of trillions of won and tens of thousands of GPUs.
That gap determines the direction of Korea's AI sovereignty. Not "building a better global AI" but "securing AI suited to Korea." Korean-language statutes, medical data, financial regulations — securing sovereignty in domains that global models cannot handle is the realistic strategy. Naver demonstrates this direction. It is attempting to break dependence on NVIDIA GPUs by building a software ecosystem based on the Intel Gaudi AI accelerator.
In August 2025, the Ministry of Science and ICT designated five sovereign AI development consortia — Naver, SK Telecom, LG, NCSoft, and Upstage — and confirmed ₩381 billion in government support. Not "full self-sufficiency" but a combination of "selective sovereignty and strategic partnership." National security and public services run on domestic models and domestic cloud; commercial applications use global infrastructure. In hardware — HBM, batteries, SMR — the position as globally indispensable supplier is consolidated; in software services, a Korea-specific layer is built.
Across the three dimensions of AI sovereignty — compute, data, model — Korea's position is not uniform. In compute, there is a distinctive duality as both consumer and supplier; in data, a structural advantage in regulatory compatibility; in the model dimension, a realistic path of selective sovereignty. That non-uniformity itself is the shape of the strategy — if one cannot dominate every dimension, finding the optimal position in each is the realistic answer.
6. A Nation at the Crossroads
Supply-chain realignment and AI sovereignty. Two currents that appear distinct on the surface converge into a single structure in Korea.
The logic of that convergence runs like this. Because Korea dominates HBM, it secures Tier 1 access to American AI chips. That Tier 1 access enables AI infrastructure investment — data centers and GPU clusters. National AI model training is built on top of that infrastructure. Those models produce sovereignty in Korean-language AI services. The structure is one in which indispensability in the hardware supply chain becomes the foundation of software sovereignty. In Chapter 4 we saw what Taiwan loses when its "silicon shield" moves to the United States. Korea's shield does not move — HBM production is in Icheon and Yongin, and that is the physical basis of Korea's negotiating leverage.
Yet the structure contains risks.
When the United States demands a forced choice. If Chip 4 turns into a blade for controlling exports to China, Korea must absorb the shock of losing the China market, which accounts for 51.7 percent of its semiconductor exports. In Chapter 7 we saw the Netherlands flip ASML's export switch at American demand. Korea has not flipped the switch — not yet. And that "not yet" is the strategic space.
When China achieves self-sufficiency. In August 2023, Huawei launched the Mate 60 Pro without prior announcement. Teardown analysis found a 7-nanometer-class chip produced by SMIC. China had done what the United States believed was impossible. On the very day U.S. Secretary of Commerce Gina Raimondo was visiting Beijing, Huawei launched the phone without any notice. Chinese internet users circulated a meme mocking Raimondo as "the unofficial brand ambassador for the Mate 60 Pro." If China succeeds in developing HBM independently, Korea's indispensability is eroded. Current analysis says "why Huawei cannot catch NVIDIA." But that analysis is a variable that could change within a decade.
When AI infrastructure dependence deepens. AWS, Azure, and GCP expanding investment in Korea is a double-edged sword. Access to the world's best AI infrastructure is gained, but the paradox arises that control over data and AI workloads transfers to foreign companies. Naver Cloud, KT Cloud, NHN Cloud — domestic operators hold limited market share. The structure could be one where infrastructure visibly expands while actual control contracts.
Despite these three risks, Korea's position holds structural advantages. Dual membership in both IPEF (U.S.-led) and RCEP (which includes China). A multiple-lane strategy that participates cautiously in Chip 4 while exporting more than half its semiconductors to the China sphere. Data governance that has received the EU adequacy decision. This dual participation is not contradiction — it is the maintenance of strategic options.
The numbers show this strategy in execution. In 2023, the same year Korea's trade balance with China turned negative for the first time (-$18 billion), the trade balance with the United States recorded an all-time high surplus (+$45 billion). China's share fell to 19.7 percent — below 20 percent for the first time in 20 years.
The structure of semiconductor exports has also shifted. In 2020, 62 percent of Korea's semiconductor exports were directed to the China sphere; by 2024, that figure had fallen to 51.7 percent. The significant rise in Taiwan's share is due to the new routing of HBM for NVIDIA passing through TSMC. The trade map is already being reconfigured. Diversification is not a slogan — it is appearing in the data.
₩600 trillion for the Yongin Semiconductor Cluster, ₩450 trillion from Samsung over five years, ₩622 trillion for the Gyeonggi Semiconductor Mega Cluster. Simultaneously, $5 billion for the SK–AWS Ulsan Data Center, $35 billion for the South Jeolla data center complex. Semiconductors, AI infrastructure, and energy are being connected into a single investment chain. At each link of that chain, Korea is supplier and consumer, partner and competitor.
The nation standing at the crossroads of supply chains holds the most options. A crossroads is dangerous — a collision can come from any direction. But leaving the crossroads means losing the options.
Korea's strategy is not to leave the crossroads. To remain — but not to be fixed to any single direction. Between the United States and China, between subsidies and the market, between hardware sovereignty and software dependence. In Chapter 3 we saw Singapore present the supply-chain critical node as the second condition of indispensability. Korea stands not on a single node but on multiple intersections simultaneously. HBM as node, batteries as node, data governance as node, and AI infrastructure as node.
The empty cleanroom in Texas is a warning. Supply-chain realignment is not completed by subsidies. SK On's factory in Georgia is evidence. A preemptive decision creates policy dividends. India's 1.4 billion people are a variable. A new gravitational center is changing the terrain itself. The three dimensions of AI sovereignty are challenges. None can be dominated by Korea alone — but the optimal position in each dimension constructs the whole strategy.
At the point where those intersections overlap, the strategy of the in-between lives.