Anyone who has bought a graphics card knows that the price of this item is neither decided by you nor by the company that makes it. For the GDDR7 used in consumer-grade graphics cards, Samsung is one of the main suppliers; for the high-bandwidth memory HBM packed into AI servers, SK Hynix is the undisputed leader. For every AI-generated image you find online, more than 80% of the memory cost behind the computing power flows into the pockets of two South Korean companies. This number is not a peak from any single year, but the standard profit distribution of this industry.
Korean chips are the foundational material of every device you and I use.
On June 29, South Korean President Lee Jae-myung convened a press conference at the Blue House called the "National Report on Three Mega Projects." Samsung Chairman Lee Jae-yong and SK Group Chairman Chey Tae-won sat on stage, introduced by the host as "national heroes" in front of a national television audience. The event was heavy on formality; the way the cameras framed the two businessmen felt more like filming a state ceremony than a corporate investment announcement.
This title is not given casually in South Korea. It typically appears in the context of national crises or major events; the last similar moment was during the 1998 IMF financial crisis, when ordinary citizens queued up to donate gold jewelry to save the country. Applying it to the heads of two corporations felt awkward to many Koreans, and the opposition party directly criticized it as "government-administered governance."
But if you look at the scale and structure of this account, you will find that beneath the awkwardness lies something worth thinking about more carefully.
25 Trillion RMB: Not a Single Year's Spending
First, let's get the numbers right.
4,700 trillion Korean won, roughly 25 trillion RMB, means this is a cumulative planned total spread across multiple years and multiple projects, not next year's fiscal budget. On the semiconductor side, Samsung plus SK Hynix will invest a combined 800 trillion won to build four factories in North Jeolla Province and South Jeolla Province in southwestern South Korea. On the AI data center side, through public-private cooperation with a total scale of 18.4GW, the first phase involves 3 companies investing 550 trillion won to build 8GW, with SK Group selected sites like Ulsan. The site selection itself is deliberate—the southwest and Ulsan are regions with relatively mature industrial bases, not starting from scratch.
What does 8GW mean? Based on the 500MW cap for a single hyperscale data center cluster, 8GW is roughly equivalent to 16 such clusters starting construction simultaneously. This shows South Korea is not betting on "having a few data centers" but on achieving scale-level presence in computing infrastructure. Once such scale materializes, it's not about single-point technological leadership but about creating barriers through density itself.
For comparison, the U.S. CHIPS Act is about $52 billion; China's National Integrated Circuit Industry Investment Fund Phase III, the European Chips Act, and Japan's subsidies for TSMC's Kumamoto plant—every country is making heavy bets in this round. What makes South Korea's card special is not that it's "also big," but that it simultaneously bets on three directions that are interlocking.
Samsung and SK: Rivals in Daily Life, Now Seated Together
On the HBM track, Samsung and SK Hynix have fought for years, and it hasn't been easy. The two companies watch each other closely on technology paths, customer resources, and yield rates, chasing each other relentlessly. The combined share of Korean companies exceeding 80% in the HBM market is the result of the two companies genuinely pushing each other forward. This competitive relationship truly drives technological iteration; it's not for show.
Now they are called to the same table, backed by the president's endorsement, and together they announce plant construction in the same region in the southwest.
There is something subtle here. The competitive relationship hasn't disappeared, but both companies have been anchored simultaneously into a larger narrative, becoming the "two poles" of the same national project. With the two plants located close to each other, infrastructure can be shared, and the talent pool is in the same region—this is both synergistic efficiency and a form of implicit interest bundling. This structure has appeared in Japan's semiconductor revitalization and has echoes in the operations of China's Big Fund. The question is not whether the government should participate, but whether the manner of participation ultimately builds systemic capability or just distributes money.
The opposition's criticism of "government-administered governance" points to this crack. When the government uses administrative means to marshal resources, does the whip of market competition still exist?
I believe this is the most genuine risk in this matter, and it is harder to answer than technical issues like "execution progress."
Who Are They Really Guarding Against
This money is spent on three directions: semiconductors, AI data centers, and physical AI. Calling it a "trinity" is official language, but the logic of these three directions does form a single thread: memory chips are the hardware foundation, data centers are the computing carrier, and physical AI is the final application form. South Korea has chosen a path that threads from the bottom up. The advantage is that each layer can be monetized; the risk is that if any layer lags behind, the overall narrative breaks.
In the robotics direction, South Korea's position is: it is not a leader in components. In the key component of reducers, Harmonic Drive and Nabtesco monopolize the high-end market—both are Japanese companies. In precision sensors, the U.S.'s ATI and Japan's Keyence are the standard-setters. From a component perspective, South Korea looks more like an assembler than a technology originator.
But an assembler solves "whether it can move," while a system-level player solves "whether it can judge and decide in complex environments." These two things are vastly different. South Korea's three real cards are laid out as follows: Boston Dynamics' control algorithms plus Samsung and LG's sensor integration solve the system integration problem; HBM and customized DRAM bring a unique advantage at the chip-robot intersection, where on-chip memory bandwidth directly affects the speed of a robot's real-time decision-making; Hyundai Motor and Kia's annual production line of 7 million vehicles is a verification field no one else has. A vehicle factory is not just a sales number—it is a pressure-testing platform for robots operating in real working conditions over long periods. This scale of validation opportunity cannot be replicated by most robotics companies in their labs.
This money is not betting on a breakthrough in a single component, but on whether these three cards can form a cohesive strategy.
The Pressure Felt on China's Side Is Two-Directional
Once South Korea's data centers are built as planned, two things will directly change the competitive landscape.
First, the time window narrows. CXMT is working on an alternative route for HBM, and this path is not easy—technology gaps and manufacturing yields are real barriers. South Korea's large-scale capacity expansion and technology investment this round means the time window for CXMT to catch up has narrowed further, not because CXMT isn't trying, but because the other side is accelerating. Technology gaps could originally be traded for time, but the clock on the other side has been sped up, and that equation no longer holds.
Second, computing rental costs face pressure from both sides. Chinese AI companies like SenseTime and 4Paradigm that have overseas computing needs—if they rent computing power from data centers in South Korea or deeply tied to it—the price anchoring power still lies with the supply side. Long term, South Korea's capacity expansion may lower unit costs, but in the short term, before new capacity is released, the logic of computing scarcity remains unchanged. The time lag between these two sides is where the pressure lies.
These two things are not crises that will erupt on some future day; they are baselines that move every day.
Lee Jae-myung said "three mega projects" on stage, while Lee Jae-yong and Chey Tae-won sat below, called "national heroes." This image itself is a microcosm of the whole story: one government, two long-time rivals, one enormous bet, all in the same frame. After the press conference, they still have to go back to competing. The frame cannot change market logic, but it does cement a political commitment.
The bet is big enough, the structure is tight enough, but whether it succeeds depends on whether the government's hand pushes the market forward or smothers the outlet of competition.
Betting on systemic capability—the hardest part is not spending the money, but keeping competition within it.