NextFin News - South Korea's deputy prime minister has warned that the country cannot afford to slow its artificial-intelligence buildout, framing the race as a matter of national economic survival even as Seoul wrestles with a stock market that has nearly doubled on the backs of two chipmakers and a comprehensive AI regulatory framework that took effect only in January.
Bae Kyung-hoon, who serves as both deputy prime minister and minister of science and ICT, delivered the message on September 16, 2026, as the government pressed ahead with one of the largest state-coordinated technology investment programmes ever attempted. The warning lands at a moment of maximum tension for Korea's AI bet: the benchmark KOSPI index closed at 6,691 points on September 16, up 0.96% on the day but still roughly 29% below the record 9,385.59 it touched in June, while Samsung Electronics and SK Hynix have turned the country's equity market into a concentrated wager on high-bandwidth memory.
The deputy prime minister's warning is a statement about compounding advantage as much as it is about industrial policy. Once a country has committed capital on this scale, stopping is not a neutral option: it would strand assets, cede supply-chain position to competitors, and convert today's chip-driven tax windfall into tomorrow's idle capacity. In a race where the prize goes to the builder who stays ahead, slowing down is functionally the same thing as losing.
The Stakes Behind the Warning
The message is not rhetorical padding. South Korea is in the middle of executing a national AI strategy announced in June 2026 by President Lee Jae Myung that orchestrates at least 1,350 trillion won ($880 billion) of corporate investment into semiconductors and data centres. The plan rests on a "triple axis" of chips, physical AI and data-centre capacity, and its scale is difficult to overstate: for context, South Korea's entire 2027 government budget is set at 821 trillion won ($597 billion), meaning the private-sector commitment behind the AI push is roughly 1.6 times the size of all planned government spending next year.
Within that envelope, Samsung Group and SK Group have each pledged to build two new chip fabrication sites in the country's southwest, a combined 800 trillion won ($518 billion) undertaking with their suppliers. Samsung separately disclosed in a regulatory filing that it will invest 2,450 trillion won domestically between 2026 and 2040, with 2,100 trillion won of that directed at semiconductor clusters around Pyeongtaek and Yongin. On the data-centre side, SK Group, GS Group and Naver will invest 550 trillion won to deliver 8.4 gigawatts of AI data-centre capacity by 2029, with a second phase running to 2035 adding another 10 gigawatts for a total of 18.4 gigawatts and more than 1,000 trillion won in cumulative investment.
Bae has previously described those data centres as "token factories" - infrastructure that generates AI intelligence at scale - and the government plans to secure 52,000 high-performance GPUs by 2028, scaling to 260,000 by 2030 through joint public-private investment.
"Up to now, we focused on securing resources and rebuilding foundations. From next year, execution and results must follow."
Bae said at a December 2025 briefing, setting the sequencing that makes the September warning the logical endpoint.
The September 16 warning is the logical endpoint of that sequencing. It also arrives against a backdrop of unusual fiscal strength. The semiconductor boom has already transformed Korea's public finances: the government raised its 2026 growth forecast to 3.0%, the strongest pace since 2021, and corporate tax receipts are projected to more than double to 216.7 trillion won in 2027, helping total tax revenue rise 40.7% to 584.4 trillion won. That windfall is what allows Seoul to propose an 821 trillion won budget for 2027 while still expecting the debt-to-GDP ratio to fall to 48.3% from 51.6%. The AI chip boom is not just an industrial project; it is funding the state's broader agenda - regional development, youth support, fiscal consolidation.
Why Korea Cannot Stop: The Mechanism
The reason slowing down is so dangerous is structural, not cyclical. Korea's position in AI rests on a narrow but decisive chokepoint: high-bandwidth memory. Only three companies in the world produce HBM at scale - SK Hynix, Samsung and Micron - and capacity expansion carries 12-to-18-month lead times. As of the second quarter of 2026, SK Hynix held 50% of HBM revenue market share, with Samsung at 33% and Micron at 18%, according to Counterpoint Research. SK Hynix was first to mass-produce HBM3E and has secured the majority of Nvidia's supply contracts; Samsung signed an agreement in March 2026 to supply HBM4 for AMD's MI455X platform.
This is not a commodity position that mean-reverts. HBM is a qualification business: once a chip is validated inside an accelerator platform, it tends to stay there across generations. That creates a durable revenue stream, but it also creates a trap. If Korea slows its capital spending while competitors keep building, the loss is not just next quarter's margin - it is the qualification slot in the next generation of accelerators, and with it the customer relationship that pays for the following generation. The mechanism is compounding advantage, and compounding does not pause cleanly.
The pricing power behind that mechanism is real. HBM pricing has risen an estimated 20% to 40% year-over-year as demand outstrips supply, and the three-way supply concentration gives vendors leverage that pure foundry players do not have. But pricing power is a flow, not a stock: it persists only while capacity remains tight. The moment new capacity arrives - from Korean fabs or from Chinese competitors such as YMTC's parent, which is targeting a $4.9 billion Shanghai IPO to fund memory expansion - the margin that funds the whole strategy compresses.
The fiscal arithmetic reinforces the trap. If the chip cycle turned before the diversification into physical AI and data centres was complete, the government would face the worst possible combination: falling revenue and unfinished infrastructure. That is why Bae's warning carries the weight it does. The state has tied its balance sheet to the continuation of the boom.
The Counter-Thesis: A Bubble Built on Two Stocks
The strongest argument against Seoul's urgency is that the entire national strategy rests on a market that has become dangerously concentrated. Samsung Electronics has gained roughly 111% year-to-date as of September 16, and the KOSPI itself is up nearly 94% over the past year - but the breadth underneath is thin. In May, SK Hynix was up almost 200% from the start of the year and Samsung nearly 144%, while the broader market lagged. A rally powered by two memory names is exposed to a single commodity cycle, and memory is historically the most violent cycle in technology.
The volatility has already been demonstrated. The KOSPI dropped more than 20% from its June record close in early July, triggering two consecutive circuit breakers before rebounding 17.91% in a single session on July 31 to close at 6,595.45. Foreign investors offloaded a record net 148.3 trillion won ($96.7 billion) of Korean equities in the first half of 2026, driven by profit-taking on semiconductor producers and global rebalancing. A market that can fall 20% and rip 18% within weeks is not a stable foundation for a decade-long industrial strategy.
Bae has directly addressed the concentration criticism. In May, responding to questions about the market's narrow leadership, he said:
"While these two companies clearly have their own strengths, there is also an ecosystem of related companies that support semiconductor production."
The government's answer is the physical-AI leg of the triple axis - embedding AI in robots, vehicles and industrial systems, where Korea's manufacturing base gives it a claim to compete - plus the data-centre buildout that would create domestic demand for AI compute beyond the chip fabs.
The counter-thesis is not frivolous, and it has a quantifiable falsifying signal. If HBM pricing reverses for two consecutive quarters while SK Hynix's and Samsung's inventory days rise, the concentration risk becomes a solvency-level problem for the strategy, because the tax revenue and the equity gains would be reversing together. That is the single metric worth watching: HBM average selling prices and inventory, reported quarterly.
Regulation in the Middle of a Sprint
A second tension runs through the strategy: Korea is trying to sprint while also putting on a regulatory harness. The AI Basic Act - formally the Act on the Development of Artificial Intelligence and Establishment of Trust - took effect on January 22, 2026, making Korea the second jurisdiction after the European Union to adopt a comprehensive AI framework. It imposes transparency obligations on generative and high-impact AI, safety requirements for frontier models, and duties on both AI developers and AI-using business operators.
The design intention is to thread the needle between innovation and trust. Bae has pushed back against bubble fears directly:
"AI is not a bubble."
He has argued that high infrastructure and data costs discouraged private participation in the past and that government procurement would serve as a turning point. The risk is that a regulatory regime designed for caution lands on an industry whose economics depend on speed. The EU AI Act's main high-risk compliance milestones are not due until August 2028, giving European firms a longer runway; Korea is enforcing first, betting that early clarity will attract rather than deter investment.
The bet is not obviously wrong. For chipmakers selling into US and European supply chains, a credible domestic AI governance regime can reduce friction rather than add it - it signals to customers that Korean AI systems and the data they process meet recognised standards. The danger lies at the model layer, where frontier developers face safety and transparency duties that raise the cost of experimentation just as they are trying to catch up with US and Chinese rivals.
The Geopolitical Squeeze
Beyond regulation and concentration sits a third pressure that the deputy prime minister's warning cannot fully name: Korea's position between the United States and China. Korean chipmakers depend on US equipment and design tools, on US cloud and accelerator customers, and on a Chinese market that remains a significant buyer of legacy chips. Every expansion decision is therefore a diplomatic calculation as much as a commercial one.
The US export-control regime determines which tools Korean fabs can buy and which customers they can serve. China's response - building its own memory capacity and steering procurement toward domestic suppliers - threatens the very demand that justified the Korean buildout. Korea's strategy is to run fast enough that its technological lead becomes the shield: if SK Hynix and Samsung are always a generation ahead in HBM, the qualification slot matters more than the political wind. That is a sound logic only as long as the lead holds.
What Comes Next: Three Horizons and Three Scenarios
Short term (6-12 months): The market will watch whether the KOSPI can reclaim the June highs near 9,385 and whether Samsung and SK Hynix can hold their HBM pricing. The September 16 session put the index at 6,691, up 0.96% on the day - a market that is trading the AI thesis but has already given back a meaningful chunk of its peak. The near-term trigger is quarterly HBM ASP and inventory data; a clean print keeps the rally intact, a soft one reopens the 20% drawdown question.
Medium term (1-3 years): Execution of the mega-projects. The 8.4 gigawatts of data-centre capacity due by 2029 and the 260,000 GPUs targeted for 2030 are the measurable milestones. If those slip, the "execution and results must follow" promise becomes an empty one. This is also the window in which physical AI must move from pilot to revenue - Hyundai's integration of Boston Dynamics' Atlas robots into manufacturing is the poster child the government will point to.
Long term (structural): Whether Korea converts memory dominance into a broader AI position - physical AI, domestic foundation models, and the "AI-inclusive society" Bae has described. "The benefits of AI must also go to the public," he said in May, framing the goal as an "AI-inclusive society - a society where no one is left behind in the AI era." That is the political licence the strategy needs to survive a downturn.
Three scenarios frame the path:
- Base case: HBM demand stays ahead of supply through 2027, the mega-projects deliver on schedule, and Korea holds 45% or more of the HBM market. The KOSPI grinds higher and the chip-driven tax windfall funds fiscal consolidation. Probability-weighted, this is the path Seoul is betting on.
- Upside case: Korean firms convert HBM cash flow into leadership in physical AI and domestic AI factories, diversifying beyond memory. The KOSPI reclaims and exceeds its June record, and Korea achieves its stated goal of becoming one of the world's top three AI powers.
- Downside case: HBM pricing rolls over for two consecutive quarters, Chinese capacity arrives faster than expected, and the KOSPI revisits its July lows. The fiscal windfall reverses, leaving unfinished infrastructure and a debt ratio that stops falling. This is the scenario the falsifying signal is designed to catch early.
President Lee has set the tone from the top:
"We must secure the core elements of AI faster than any other country."
The deputy prime minister's September warning makes clear what is at stake if they do not. Korea has bet its fiscal future and its market on a narrow technological lead. The bet pays only if the country keeps running - because in a compounding race, slowing down is the same thing as losing.
Explore more exclusive insights at nextfin.ai.
