(SINGAPORE 2026.3.18) At the beginning of this month, data released by South Korea’s Ministry of Science and Information and Communication Technology (ICT) showed that the country’s semiconductor exports exceeded US$20 billion (sbout S$26.8 billion) for three consecutive months. In February alone, exports surged more than 160% year-on-year, reaching a record US$25.16 billion. Chips now account for more than one-third of South Korea’s total exports, making the industry the single most powerful engine of its economy.
However, about three years ago in 2022, if one word could summarize the mood in South Korea’s semiconductor industry, it would have been despair, the Chinese financial media platform Chinese Business Strategy (华商韬略） recalled.
Global demand for electronic products was weak. Prices for memory chips had collapsed. Samsung Electronics and SK hynix were piling up unsold inventory in warehouses in Yongin and Icheon. Exports had declined for twelve straight months, and an industry once synonymous with precision and dominance suddenly appeared fragile.
Yet just a few years later, the situation reversed dramatically—from a winter-like contraction to unstoppable growth. What caused this transformation? The answer begins with a gamble—a huge, uncomfortable, almost irrational gamble, made at the very bottom of the market.
Back in 2022, before ChatGPT ignited the global AI boom, the semiconductor industry was in a slump. The personal computer and smartphone markets were saturated, and memory chips—especially DRAM—had become commoditized and cheap.
Competitors reacted predictably: cut production, reduced investment, and hoarded cash. Even the main rival of Samsung and SK hynix—US Micron Technology—announced plans to scale back operating factory capacity to historic lows.
Within Samsung and SK hynix, similar debates were underway. Finance teams advocated for survival—cut high-risk R&D, especially projects unlikely to pay off in the short term. Chief among them was HBM (High Bandwidth Memory).
HBM is notoriously difficult to manufacture. It requires vertically stacking multiple DRAM chips—like building a miniature skyscraper—and connecting them using Through-Silicon Via (TSV) technology. Thousands of tiny holes are drilled into fingernail-sized areas, each requiring exact precision. A single defect could ruin the entire wafer. The process is costly, complex, and risky.
Yet it also promised revolutionary change.
At SK hynix, CEO Kwak Noh-Jung (郭鲁正） decided not only to continue R&D but to double down. He believed traditional memory had become a red ocean market. If the company wanted a future, it had to bet on next-generation technology. Even if it took years, AI would demand fundamental innovation.
Thus, at the coldest point of the cycle, SK hynix increased HBM R&D spending by 30%, focusing on improving HBM3E yield. By 2023, it became the first company in the world capable of reliably mass-producing HBM.
Samsung Electronics followed suit. Despite losing more than US$10 billion in its semiconductor business, it pushed ahead with HBM4 development. It was a critical choice: if AI failed to materialize, these investments could have severely damaged both companies; if successful, they would dominate the future.
Within months, everything changed. ChatGPT went viral globally. Large language models required massive amounts of data. Having powerful GPU was no longer enough—memory bandwidth became crucial. Even NVIDIA’s most advanced chips, like the H100, were limited by data movement rather than computing power.
HBM was no longer optional—it became essential. Only three companies in the world could produce it at scale: SK hynix, Samsung, and Micron. The two Korean companies together controlled nearly 80% of the market.
By 2026, SK hynix’s annual HBM capacity was fully booked. New orders? Customers would have to wait until next year. What seemed like reckless spending in 2022 has now become an almost perfect forward-looking strategy.
But technology alone was not enough. To win in HBM, Korean companies adopted an unusual strategy: turning high-tech manufacturing into a service business.
HBM chips cannot simply be plugged into systems—they must be closely integrated with GPUs. This meant working side by side with customers.
Reportedly, SK hynix engineers sat next to NVIDIA teams at NVIDIA’s headquarters in Santa Clara. During the H100 development, issues like cooling and chip stacking were not solved by email or meetings—they were resolved on-site through real-time discussions.
This “embedded collaboration” model gave SK hynix a key advantage. While competitors were still debating specifications, SK hynix was already delivering usable samples. It also achieved something even more valuable: insight. Being close to the technological frontier allowed them to anticipate the next technical step—like HBM4—before the market demanded it.
This victory did not come overnight; it was the result of decades of accumulation. As early as 1983, Samsung, under founder Lee Byung-chul (李秉喆）, boldly entered the semiconductor industry. At the time, Japan dominated the global market, and South Korea had neither the technology nor the ecosystem.
The strategy was ruthless: invest relentlessly, even at massive losses. Through the Asian financial crisis, the 2008 global downturn, and multiple industry recessions, Samsung and SK hynix followed the same pattern—investing more when others pulled back.
They endured years of losses, piling up discarded wafers and forcing engineers to work overtime. Competitors exited, leaving South Korea alone. Over time, they not only survived but became the last stronghold.
By 2025, SK hynix’s annual operating profit is expected to reach 45 trillion won (about S$44 billion). Samsung’s semiconductor division set quarterly profit records, with HBM as the core driver.
What seems today like a windfall from the AI boom is actually rooted in deeper factors: the courage to bet counter-cyclically, focus on core competencies, deep integration with ecosystem-leading companies, and decades-long strategic thinking.
As Kwak Noh-Jung later reflected, the luckiest decision was actually simple: not to retreat at a critical moment. In the end, this was not a stroke of fortune, but wealth won through conviction, endurance, and an all-in bet on the future.
