1. The Weight of Eight Teeth
On December 26, 2024, Netflix released Season 2 of Squid Game simultaneously in 190 countries. 68 million views in four days. The fastest launch in the history of Netflix TV series. It displaced the previous record holder — Wednesday, at 50.1 million views — within a single day.
Eleven days in, the number reached 126.2 million views. Ranked number one simultaneously in 93 countries. The top Netflix series by total viewing hours in the second half of 2024 — 619.9 million hours.
Hwang Dong-hyuk had waited thirteen years for this moment.
In 2008, he was deep in financial difficulty. He sold family insurance policies and borrowed money. In his mind circled a story: adults playing the street games of childhood with their lives on the line. He took the screenplay to film companies and broadcasters. The answer was always the same. "Too violent." "Unrealistic." The rejections kept coming, and the screenplay was buried in a drawer. In 2019, as Netflix moved in earnest to invest in Korean content, that drawer was opened again. Filming began. During production, Hwang Dong-hyuk lost eight teeth to stress.
September 17, 2021 — a simultaneous global release. The results of Season 1 became legend. Production cost: $21.4 million (approximately ₩28 billion) — $2.38 million per episode (approximately ₩3.1 billion). Netflix's estimated content value: more than $900 million (approximately ₩1.2 trillion). Forty-two times the cost. Cumulative audience of 330 million, total viewing time exceeding 2.8 billion hours. The most-watched series in the history of humanity.
Yet Hwang Dong-hyuk received no additional compensation from Season 1's success. He later said that agreeing to produce Season 2 was "a negotiating card to receive fair compensation for the success." The production cost of Season 2 rose to approximately ₩100 billion (roughly $68.6 million) — more than three times Season 1. That difference shows, in numbers, where Hwang Dong-hyuk's negotiating leverage stood — and how far the market value of K-culture had risen.
One year earlier, in February 2020, Bong Joon-ho's Parasite won Best Picture at the Academy Awards — the first non-English-language film to do so in the Academy's 92-year history. Production cost: ₩15.5 billion (approximately $13 million). Global box office: $263 million (approximately ₩342 billion). Bong Joon-ho said in his acceptance speech: "Once you overcome the one-inch barrier of subtitles, you will be introduced to so many more amazing films." After that one inch was crossed, Netflix decided to invest $2.5 billion (approximately ₩3.25 trillion) in Korea. Parasite prepared the soil. Squid Game grew in it.
Thirteen years of rejection, the cost of eight teeth, and the all-time record for humanity. This trajectory holds, simultaneously, the structure that makes K-culture possible and the contradictions embedded within that structure. Extreme concentration, extreme training, extreme endurance. And at the end of it, the gap between those who capture the value and those who do not.
2. Soft Power from a Factory
K-culture comes from a factory. That is not a romantic expression. It is an industrial fact.
Gangnam-gu, Seoul. SM Entertainment's basement rehearsal room. A sixteen-year-old trainee's day is filled with eight hours of dance, two hours of vocals, one hour of foreign-language class, and one hour of physical conditioning. After twelve or more hours of daily routine, the trainees return to the dormitory and write in their journals. With no certainty about whether they will debut or be cut.
Average training period: three to five years. At the major agencies, five to eight years is not unusual. The monthly cost borne by an agency per trainee averages approximately ₩1.6 million. JYP Entertainment's trainee development costs grew 30 percent from ₩850 million in 2023 to ₩1.12 billion in 2024.
The attrition rate in this system is brutal. A 2023 survey by the Korea Creative Content Agency found an annual voluntary withdrawal rate of 34.4 percent. Of all trainees, only 10 to 20 percent actually reach debut. The trainee count, which stood at 1,895 in 2020, fell 38.3 percent to 1,170 in 2022. Shifts in Generation Z values, growing attention to mental health, and awareness of future uncertainty are reducing the supply itself.
The output of this factory shook the world. April 15, 2023. Indio, California. Coachella Valley. BLACKPINK took the main stage. The first Korean artist to headline Coachella in its 24-year history, the first Asian headliner, the first female group to headline. Four years after performing on the secondary stage in 2019. The move from sub-stage to headliner was visible proof that K-pop had crossed from the margins of global popular music to the mainstream.
The numbers confirm that move. Korea's cultural content exports in 2023: approximately $13.1 billion (approximately ₩17.5 trillion). Up 2.6 times from $5 billion (approximately ₩6.5 trillion) in 2015, over eight years. Games were the largest single category at $8.2 billion (approximately ₩10.7 trillion), with broadcasting and drama at $1.5 billion (approximately ₩2 trillion), music at $900 million (approximately ₩1.2 trillion), and webtoons at $800 million (approximately ₩1.04 trillion). Most attention goes to K-pop alone, but the actual largest category of content exports is games. Krafton earned 95 percent of its 2024 revenue of ₩2.1 trillion from overseas — effectively a global company. Nexon became the first Korean gaming company to break ₩4 trillion in annual revenue.
The combined revenue of the Big 4 entertainment agencies reached ₩4 trillion (approximately $3 billion). HYBE alone posted ₩2.2 trillion — the first K-pop agency to cross ₩2 trillion. Yet even HYBE exposed its vulnerabilities. When former BTS members entered military service, operating profit fell 38 percent in 2024. The structural risk of the K-pop industry — excessive dependence on specific icons — was laid bare.
Netflix announced it would invest $2.5 billion (approximately ₩3.3 trillion) in Korean content for 2023–2025. After Season 1 of Squid Game, the global subscription revenue generated by Korean dramas on Netflix reached $3.4 billion (approximately ₩4.4 trillion). Among non-English content, Korea ranks first with 8.71 percent of total viewing hours.
Webtoons erased borders too. Lore Olympus, posted on Naver Webtoon's Canvas platform by New Zealand creator Rachel Smythe, swept all three major American comics awards — Eisner, Harvey, and Ringo — simultaneously in 2022. Cumulative views: 1.8 billion. An animated series has been confirmed for Amazon Prime Video. This is not the work of a Korean creator. It is a New Zealand creator drawing Greek mythology on a Korean platform. Naver Webtoon had 170 million global monthly active users in 2024 and a 70.5 percent share of the North American webtoon market. K-webtoon is not content export — it is platform export.
This content export pulls consumer goods along with it. In 2024, at a Walmart in Texas, Nongshim Shin Ramyun moved from the Asian food aisle to the general food section. Nationwide placement across all 4,692 Walmart locations. A 25.4 percent share of the U.S. instant noodle market, ranking second. In the same year, CJ CheilJedang's Bibigo mandu secured the top position in the North American frozen dumpling market — a margin more than three times the second-place brand. Overseas food sales of ₩5.58 trillion exceeded domestic sales for the first time. Academic research has validated the effect: "a 1 percent increase in cultural product exports generates a 0.136 percent increase in consumer goods exports" — and it was operating in reality.
In Chapter 3 we saw Singapore make the reproduction of human capital the fourth condition of indispensability. The K-culture trainee system is an extreme form of human capital reproduction. From a pipeline of 1,170 trainees, 34.4 percent are cut each year, and the surviving 10 to 20 percent compete for global audiences. This structure produced Squid Game, BLACKPINK, and Naver Webtoon.
But this soft power has a fatal limit. It is replaceable.
China's short-form drama market surpassed ¥50.4 billion in 2024. Approximately 30 percent of international K-drama fans also regularly watch C-dramas. Chinese-language dramas on Korean streaming platforms more than doubled in three years. In Chapter 11 we saw Samsung Foundry's yield problems. Just as Samsung Foundry was replaceable by TSMC, K-culture is replaceable by C-dramas and Japanese animation. Replacing a semiconductor requires decades and hundreds of billions of dollars. Replacing a drama requires one season.
The hallyu ban proved this vulnerability. After the THAAD deployment in 2016, China took the Korean Wave hostage. Economic losses of ₩22 trillion ($15.3 billion). A 48.3 percent drop in Chinese tourists. A blockade lasting effectively nine to ten years. Soft power as hostage rather than leverage — a textbook case.
Japan's Cool Japan strategy offers a cautionary lesson. This national project, pursued by the Ministry of Economy, Trade and Industry in the 2010s, posted pre-tax losses of more than ¥10 billion (approximately $88.9 million) within five years. The government directly planned content and bureaucrats set direction. Japan overlooked the fact that successful cultural phenomena rise organically from below. Korea has not yet repeated this mistake — the government lays infrastructure, and creation is left to the private sector. But the moment K-culture is directly mobilized as a diplomatic instrument, it risks the same backlash Japan faced.
K-culture is not Taiwan's silicon shield. In Chapter 4 we saw TSMC's silicon shield — geopolitical deterrence built from irreplaceability. K-culture cannot serve as that kind of shield. It is a "reputation asset" that generates familiarity and goodwill — not hard leverage that deters external pressure in a crisis.
3. The Apologizing Chaebol
October 8, 2024. Jun Young-hyun, vice chairman and head of Samsung Electronics' DS Division, sent an unusual message of apology. It was an official communication addressed to employees, investors, and customers.
"We have caused concern to our customers, investors, and employees with results that fell short of market expectations."
It was unprecedented for the head of the semiconductor division at Korea's largest company to publicly apologize for underperformance. The core of chaebol culture is the myth of infallibility. Admitting failure is a crack in that myth. Samsung Electronics alone accounts for approximately 21 percent of Korea's total exports in 2024. Samsung's fortunes are Korea's economic fortunes. An apology from the semiconductor head of that company carries a weight beyond simple underperformance.
Behind the apology was Samsung's elimination from the HBM competition. Samsung Electronics' HBM3E failed NVIDIA's qualification tests throughout 2024 due to thermal and power-consumption problems. During the same period, SK hynix was the sole supplier of HBM3E for NVIDIA's H100 and H200. In June 2024, Jensen Huang, NVIDIA's CEO, said publicly that he hoped "Samsung's testing would conclude quickly." By October, supply approval had still not been granted.
The foundry situation was more serious. Samsung declared it had achieved mass production of a 3-nanometer GAA process — the first in the world — in 2022, but could not escape a yield of 50 percent. TSMC's 3-nanometer yield exceeded 90 percent in the same period. The gap was not simply a number — it was the collapse of trust.
Google moved its Tensor chip to TSMC starting from the fifth generation. Qualcomm placed an exclusive order for its next-generation flagship chip on TSMC's 3-nanometer process. The Samsung Galaxy S25 would carry a Qualcomm chip made by TSMC instead of its own Exynos. The situation: unable to put its own chip in its own smartphone. The foundry division posted an estimated annual loss of ₩4 to 5 trillion in 2024. More than 50 percent of production lines were idled.
TSMC founder Morris Chang delivered a direct blow at a forum in December 2024: "The fundamental reason Samsung and Intel have failed in foundry is the conflict-of-interest structure. They must manufacture their own products while simultaneously manufacturing the products of their competitors." In Chapter 4 we saw Morris Chang found TSMC at age 56. The pure-play foundry he designed — a structure that does not compete with customers — is the wall Samsung cannot clear.
After Jun Young-hyun's apology, Samsung separated the manufacturing organizations of memory and foundry again. It reversed the integration it had carried out three years earlier. That was an admission that the integration had failed. An AI center was established within the DS Division, and personnel criteria were reset from management executive-centered to technology executive-centered. The head of the foundry division was replaced.
A quiet restructuring also began. In the autumn of 2024, the DS Division examined voluntary-retirement packages targeting senior employees. The official position was "unfounded," but the industry knew. The number of Samsung Electronics employees aged 40 and above had already surpassed those in their 20s and below. The terms discussed involved a severance of approximately ₩400 million — base retirement pay plus four months' salary. The organization that the AI era demands cannot be built from the experience of process specialists who joined in the 1990s alone. Preferred competencies in job postings were shifting to AI and machine learning, data analysis, and software engineering. The workforce structure of the semiconductor empire was quietly being rebuilt from within.
Within the same chaebol structure, SK hynix produced the opposite result.
In 2024, a dynamic emerged in which NVIDIA was pressing SK hynix for more HBM. The industry called this Super Eul — the "super subordinate." SK hynix, long in the subordinate position as a component supplier, held negotiating leverage at the narrowest bottleneck of the AI boom. In Chapter 11 we saw SK hynix's overwhelming HBM dominance — that market position was proven in numbers. 2024 revenue: ₩66.19 trillion. Operating profit: ₩23.47 trillion — both the highest figures in the company's history. Year-on-year revenue growth of 102 percent. In Chapter 7 we saw ASML build a globally dominant position in EUV lithography equipment. "Can a single company create a nation's indispensability?" — SK hynix provided the closest answer to that question with HBM.
The difference was the direction of concentration. SK hynix made preemptive investments in HBM technology from 2020 to 2022. While Samsung was simultaneously pursuing HBM, foundry, AI accelerators, and on-device AI, SK hynix concentrated its resources on the single point of HBM. The chaebol-style "big bet" paid off on one side and missed on the other. The same structure produces divergent outcomes. The direction of concentration decides the fate.
Hyundai Motor Group chose another direction. It is jointly building an AI factory with NVIDIA at a scale of 50,000 GPUs, and internalized robotics AI through Boston Dynamics. Of its ₩125.2 trillion in domestic investment for 2026–2030, more than 70 percent is allocated to AI, robotics, and autonomous driving. In Chapter 9 we saw the UAE use Mubadala and ADIA to acquire stakes in future technologies in advance. Samsung's ₩450 trillion investment plan and Hyundai's ₩125.2 trillion allocation reveal a structure unique to Korea — the chaebol effectively acting as quasi-sovereign wealth funds. Rather than the state directly managing capital, the chaebol perform that function. Even within the same chaebol model, where the concentration is directed determines the position in the AI era.
4. Those Who Leave, Those Who Remain
Korea's AI talent is leaving.
As of 2024, Korea's net outflow of AI professionals stands at minus 0.36 per 10,000 people — ranking 35th among the 38 OECD countries. Compared with Luxembourg at plus 8.92, Germany at plus 2.13, and the United States at plus 1.07, Korea is a net exporter of AI talent. Of AI undergraduates, 38.6 percent go on to graduate programs abroad. The number of Korean science and engineering doctoral researchers in the United States doubled in eleven years — from 9,000 in 2010 to 18,000 in 2021.
Compensation is part of the reason. The AI wage premium — the wage surplus over non-AI fields — is 6 percent in Korea. In the United States it is 25 percent, in Canada 18 percent, in the United Kingdom and France 15 percent. An AI faculty member's annual salary in Seoul is approximately ₩100 million. Overseas offers exceed ₩350 million. Meta's Mark Zuckerberg is reported to have personally offered contract bonuses of up to $100 million (approximately ₩137 billion) to top AI researchers in 2025.
But there are reasons deeper than salary.
The keywords that Korean AI researchers repeat in online communities fall into three categories. Research autonomy — "In Korea, publishing a single paper requires a team leader's approval." Hierarchical culture — "In a seniority-based structure, ideas are adopted in order of age." Infrastructure — "Running an experiment that needs an A100 cluster takes six months just to get the budget approved."
The more capable, the more likely to leave. Korea's research workforce is projected to fall by more than 20 percent by 2040. This is the structure of Korea's AI talent drain.
Those who have not left are fighting a different war.
Naver released HyperCLOVA X in August 2023. 204 billion parameters — surpassing ChatGPT's (GPT-3) 175 billion. Its training volume in Korean data was 6,500 times that of ChatGPT. More than ₩1 trillion invested in AI technology over five years. More than 600,000 servers running at Asia's largest data center.
The technology was built. But the market did not follow. Korea had built the AI best at the Korean language in the world — and people were still using ChatGPT. The industry's assessment was blunt: "The technology exists, but there is no killer app." The strength of Korean-language specialization is simultaneously the weakness of limited competitiveness outside the Korean-language market.
Naver pivoted to a "sovereign AI" strategy. "AI that understands Korea's language, culture, law, and institutions cannot be built by foreign companies. That is our reason for being." It is pursuing cooperation on locally specialized AI models with Middle Eastern countries including the UAE and Saudi Arabia, attempting to turn the concept of sovereign AI itself into an export product. It is a strategy to flip the limitation — that Korean-language-optimized AI is weak outside the Korean-language market — into the rationale that "every country needs an AI in its own language."
AI startups found their own modes of survival. In June 2024, Wrtn, called Korea's largest generative AI startup, conducted a large-scale involuntary separation. Target employees were brought into the office on a Friday morning and notified the same day. In the same year, Upstage attracted a strategic investment of $74 million (approximately ₩96 billion) from Amazon, seeking survival through a different route. One cut headcount; the other secured a global partner. Upstage CEO Kim Sung-hoon's strategy was explicit: "We do not fight head-on against Naver and Kakao. We take the segments they are not looking at first, and from there we go global."
Areas where the chaebol has not yet arrived, or has no intention of entering. That gap is the survival space of Korea's AI startups. Total venture capital investment in Korea in 2024: ₩11.9 trillion. AI sector investment surged 75.1 percent year on year. Rebellions focused on AI semiconductors (NPUs) and raised a Series C of $253 million (approximately ₩329 billion) — the largest ever in Korea's AI hardware sector. Yet Korea ranks ninth globally in private AI investment ($1.39 billion, approximately ₩1.8 trillion). The gap in scale has widened.
In Chapter 12 we saw the AI Basic Act delayed three and a half years. While regulatory rigidity slowed the AI transition, startups had to find their way between the shadow of the chaebol and the gaps in regulation.
Korea has 13 unicorns. Population of 52 million, GDP approximately seven times that of Israel. Israel has 23 unicorns. Population of 9.4 million. Israel's Unit 8200 trains approximately 1,000 young people each year in cyberwarfare and AI, and alumni account for 80 percent of Israel's cybersecurity company founders. Korea's mandatory military service is an 18–21 month career gap. After discharge, the goal is employment at Samsung, SK, or Hyundai. Whether military service becomes a startup incubator or a career gap is a question of institutional design.
Chaebol jobs account for only about 10 percent of total employment, but highly educated young people spend years cycling through chwijun — job preparation — for that 10 percent. A 2024 survey found that the factor most influencing willingness to take small and mid-sized enterprise jobs was "subjective norms" — social perception. "How others will see it" weighs more heavily than personal preference. The very social structure in which Samsung Electronics' "plan to hire 60,000 over five years" becomes news is itself evidence of chaebol dependency.
While the twenty-somethings who could be founding AI startups spend two to three years preparing for chaebol entrance exams, peers in Israel and the United States have already grown their startups to Series A. In a structure where the average salary at small and mid-sized enterprises stands at 63 percent of large chaebol companies, entrepreneurship is difficult to make economically rational. This is the fundamental bottleneck of Korea's AI ecosystem.
5. Concentration — A Double-Edged Blade
K-culture and the chaebol emerged from the same root. Extreme concentration, systematic training, centralized decision-making. The K-pop trainee system grinds trainees for five to eight years to produce global stars. The chaebol channeled capital in concentrated bursts over decades to build world-class manufacturing companies. Samsung Electronics' 2024 R&D investment of $24.1 billion (approximately ₩35 trillion) ranks second among global companies — ahead of both Google and Microsoft. The concentration level at which the top four chaebol account for 40.8 percent of GDP has no parallel in the world.
Is this concentration an asset or a liability in the AI era?
The answer is "both, simultaneously."
SK hynix concentrating its investment in HBM to become NVIDIA's core supplier is an instance of concentration operating as an asset. Samsung pursuing HBM, foundry, AI accelerators, and on-device AI simultaneously — and missing the timing — is an instance of concentration diluted into dispersal, becoming a liability. The same chaebol structure, the same owner management, the same vertical integration. Depending on the direction of concentration, the outcomes diverge into opposites.
The K-pop trainee annual attrition rate of 34.4 percent. The survival war of Korea's AI startups. Between these two numbers lies a structural resemblance. A small number of survivors achieve global success, and the rest exit the system. The 10 to 20 percent of 1,170 K-pop trainees who debut — and the AI startups finding niches to survive under the chaebol's shadow. The top of the pyramid is brilliant, but the pyramid itself is built by shaving down the base.
The chaebol's vertical integration is overwhelming in AI hardware. The figures seen in Chapter 11 confirm this — Samsung and SK hynix combined hold the dominant global majority in both DRAM and HBM markets.
But in AI software and models, the chaebol's organizational culture — owner-centered decision-making, zero tolerance for failure, seniority hierarchy — blocks innovation. What the AI era demands is distributed decision-making capable of running thousands of small experiments in parallel. The chaebol's single-owner system collides head-on with that demand.
AI is intersecting these two worlds. HYBE launched Syndi8 in June 2024 — a virtual pop group composed entirely of AI voices, using technology from its subsidiary Supertone. It succeeded in recording an artist's songs in six languages using AI. But when an AI-generated image was used in TXT's comeback trailer, fans pushed back immediately. "Low-grade AI" was the criticism leveled. What K-pop fandom demands at its core is "authenticity" — the blood and sweat of the artist. AI raises productivity, but it cannot replace that authenticity. At least not yet.
At the same time, AI translation and dubbing are accelerating the global spread of K-content. The global AI video dubbing market is projected to grow from $31.5 million (approximately ₩41 billion) in 2024 to $397 million (approximately ₩516 billion) in 2032. AI-dubbed content viewing time is growing 78 percent, and average viewing duration increases 60 percent. But this technology cuts both ways. As AI translation lowers the language barrier for K-dramas, it globalizes C-dramas and Japanese animation at the same speed. The global distribution capacity that was K-culture's exclusive advantage becomes universal.
The soft power of K-culture and the hard power of the chaebol. Both were born from the same structure — extreme concentration — and they share the same limits. K-culture is replaceable. The chaebol is at a disadvantage in innovation diversity. Just as the K-pop trainee pipeline is drying up, the chaebol's AI talent pipeline is draining overseas.
Yet the history of concentration has not entirely ended. There are cases of successfully redirecting the object of concentration.
HYBE's Bang Si-hyuk turned fandom into a platform. Weverse serves 245 countries and achieved profitability in 2025. A company that once depended on BTS for 95 percent of revenue recorded its highest-ever revenue during BTS's military service period. The share of BTS in HYBE's revenue fell to below 20 percent in 2024. From dependence on a single act to platform diversification — the object of concentration shifted from idols to infrastructure. The 82-concert, 34-city world tour scheduled after all BTS members complete their service in June 2025 is a bonus layered on top of that foundation.
Naver Webtoon moved beyond exporting Korean content to become the platform of choice for creators worldwide. The global webtoon market is projected to grow from $10.85 billion (approximately ₩14.1 trillion) in 2025 to $60.25 billion (approximately ₩78.3 trillion) in 2031. Naver and Kakao hold 67.5 percent of that market. Kakao's Piccoma ranked first in Japanese comics app revenue and surpassed ¥100 billion in annual transaction volume. Korea mounted a counter-offensive into the Japanese manga market through Korean platforms.
Concentration is Korea's DNA. Whether K-pop or semiconductors — going all-in on one thing and reaching the top of the world. For that DNA to function in the AI era, the object of concentration must be chosen correctly. SK hynix made the right choice in HBM. Samsung Foundry did not. K-pop built a global fandom, but that fandom offered no shield against the hallyu ban.
K-power comes from concentration. The chaebol's shadow comes from concentration. Power and shadow are products of the same light source. Hwang Dong-hyuk concentrated for thirteen years to make humanity's all-time top series — and lost eight teeth. SK hynix concentrated on HBM to become NVIDIA's core supplier; Samsung dispersed across multiple fronts and was pushed back on both. 1,170 trainees endure a 34.4 percent attrition rate and dance toward the world, while AI researchers face a 6 percent wage premium and board flights to Silicon Valley.
Where the light source is directed determines how Korea stands in the between in the AI era.
K-pop has demonstrated: innovation draws in capital, capital creates disparity, and disparity tests institutions. The trainee system systematized by SM, YG, and JYP changed the grammar of the global entertainment industry. That innovation consolidated into ₩4 trillion in capital across the Big 4 agencies. But the pyramid structure that consolidation produced — the 80 to 90 percent of 1,170 trainees who are cut, the soft power taken hostage by the hallyu ban, the vulnerability that sees 38 percent of operating profit evaporate when BTS enters military service — has already generated tensions that demand institutional response.
Technological innovation → Concentration of capital → Social instability → Institutional redesign
The chaebol follows the same path. SK hynix's innovation of concentrated HBM investment drew capital in, and the four major chaebol's revenue concentrated to 40.8 percent of GDP. That concentration is now generating tensions in the form of AI talent drain and a contracting startup ecosystem. Laws protecting K-pop trainees have been discussed, and the AI Basic Act has passed — but chaebol governance reform and measures to halt talent drain remain unfinished. Maintaining the DNA of concentration while correctly redirecting its object — that is the task K-culture and the chaebol must complete for the AI era.
The next chapter examines the structural exclusion that determines who, in Korea's economy of concentration, is pushed outside the system entirely.