The Two Koreas That Will Build the Next Decade of AI

The Number That Should Make Everyone in Tech Pause

Samsung Electronics and SK Hynix are preparing to post something that sounds almost like satire: combined quarterly operating profit of roughly 210 trillion won in the fourth quarter of 2026. That is about 150 billion dollars. It is also larger than the annual GDP of countries like New Zealand, Uganda, and Sri Lanka — economies that together account for roughly 40 million people living on what these two Korean companies are generating in a single quarter from selling bits to silicon.

The third quarter is already tracking toward 187 trillion won in combined profit. Two quarters ago, that number was 150 trillion. Each step up is roughly 30 trillion won added in a span of weeks, not years. The drivers are straightforward but historically significant. AI server demand has turned memory from a commodity into a toll road. Every model that trains, every inference pipeline that scales, every data center that expands needs high-bandwidth memory in quantities and at prices that the industry has never seen sustained at this level. HBM4 is the latest product to arrive, and it is arriving hungry.

What the Numbers Actually Mean for the Global Chip Supply Chain

The conventional narrative around semiconductor earnings is always about volume and price. But the story these Korean numbers tell is one about concentration. A handful of factories in South Korea — predominantly in cities like Hwaseong and Icheon for Samsung, and the Sinhung facility for SK Hynix — are producing enough advanced memory to power roughly the entire frontier of artificial intelligence infrastructure worldwide. That is a bottleneck so narrow it should be a strategic concern for any government that has publicly stated ambitions in AI, defense, or computing sovereignty.

The math behind the bottleneck is stark. According to market research firm TrendForce, enterprise SSD demand alone is projected to grow more than 80 percent year over year in 2026. HBM4 average selling prices are expected to more than double by next year, even as production ramps slowly. Samsung and SK Hynix together hold near-monopoly positions in the HBM segment. Micron is the only other player at meaningful scale, and it is two to three generations behind in packaging density. When you combine that with the fact that memory inventories across the industry have fallen below ten days — a historically low threshold that leaves zero cushion against disruption — the picture becomes one of a supply chain running on tightness that is barely sustainable.

This is not a temporary spike. It is a structural rewiring of who profits from the AI era. For the first time, the money is flowing upstream. The companies that design chips, fabricate nodes, and package memory are extracting margins that would have been unimaginable even three years ago. NVIDIA’s gross margins remain impressive, but they are built on a product that cannot exist without the memory these two Korean firms are shipping. Every A100, every H100, every B100 and the next generation of Rubin-class GPUs is a silicon platform awaiting HBM stacks that Samsung and SK Hynix alone can deliver at volume.

The China Question Nobody Is Solving Easily

The geopolitical dimension is arguably more consequential than the pricing story. China has spent the better part of a decade trying to build an autonomous semiconductor ecosystem. The results have been mixed at best. SMIC can now produce chips at seven nanometers, but it cannot access the most advanced EUV lithography equipment, and it certainly cannot replicate the HBM packaging expertise that SK Hynix and Samsung have refined over more than a decade of iterative yield improvement.

American export controls have tightened around this gap intentionally. The restrictions target not just the equipment needed to make advanced logic chips, but also the memory products most critical to AI training. HBM4, by design, sits at the intersection of those restrictions. It is a product that China cannot easily produce and cannot legally import in quantities large enough to fuel a competitive AI infrastructure buildout. The effect is asymmetric: it slows Chinese advances while rewarding the very companies — Samsung and SK Hynix — that are subject to American jurisdiction and willing to comply.

That compliance is not without cost. South Korea’s economic dependence on China remains enormous. China is Samsung’s largest export market for foundry services and one of its biggest markets for consumer electronics. A full rupture would be devastating. But the current trajectory suggests South Korea is choosing alignment with Washington over diversification toward Beijing, at least when it comes to the most advanced technology segments. The memory supercycle gives Seoul leverage it has not had in years: the world needs what these companies produce, and the companies themselves are recording quarterly profits that fund R&D, expand capacity, and deepen the technological gap with any competitor.

The Asymmetric Pain Inside Samsung Itself

There is another layer to this story that often goes unnoticed. Samsung is not just a memory company. It is also a consumer electronics giant, and its DX division — which covers smartphones, TVs, home appliances, and display panels — is bleeding. The company is expected to post an operating loss of roughly 1.4 trillion won in that division in the fourth quarter alone. The loss has been persistent. In the third quarter, it was already above 1.1 trillion won. Part of the reason is simple cost pressure: memory prices are rising, and those costs are cascading into the price of every smartphone, every tablet, every device that contains DRAM and NAND. Counterpoint Research has reported that global smartphone prices rose roughly 15 percent in 2026, largely because manufacturers passed component costs onto consumers.

This creates a strange internal dynamic at Samsung. The company is simultaneously profiting from and suffering under the same supercycle. The DS division funds the losses in DX. The memory business subsidizes the consumer division. If the supercycle were to soften — and it will, eventually — Samsung would face a double hit: declining memory margins and an unprofitable hardware business that has been kept alive by cross-subsidization. For investors, this duality is the kind of nuance that matters.

Who Wins, Who Loses, and What Comes Next

The winners in this environment are obvious: Samsung, SK Hynix, and the ecosystems that supply them. Taiwan Semiconductor, which fabricates the logic chips that pair with this memory, is also benefiting indirectly. Equipment suppliers to both Samsung and TSMC are seeing order books extend further out than anyone predicted even two years ago. Micron, despite being behind in HBM, is catching a ride on elevated DRAM and NAND prices that lift its entire portfolio.

The losers are less celebrated but equally real. Cloud providers and AI model developers face increasing costs for the hardware that powers their infrastructure. Every dollar that goes to memory pricing is a dollar that does not go to training more parameters or serving more requests. Chinese AI companies, constrained by export controls, face a harder path to parity. Consumer electronics manufacturers operate on thinner margins precisely when their input costs are highest. And governments that assumed chip sovereignty was achievable without controlling memory production are waking up to a reality that offers no shortcut around the South Korean bottleneck.

TrendForce projects that DRAM contract prices will rise another 10 to 15 percent in the fourth quarter, with NAND climbing 15 to 20 percent. Enterprise SSD pricing is expected to accelerate even further. KB Securities analyst Kim Dong-won noted in a recent report that there are no signs of order reductions from major customers for HBM and high-performance DRAM, and that inventory remains at historically low levels. The implication is that the supercycle has momentum that will carry well into 2027.

What makes this cycle different from previous memory upswings is its duration and its demand driver. Past supercycles were fueled by cyclical upgrades in PCs and smartphones. This one is anchored in a structural shift: AI infrastructure is not a consumer discretionary purchase. It is a capability that governments and enterprises are building because they believe they need it to remain competitive. That belief is not going away soon, and neither is the hardware that requires it.

Samsung and SK Hynix are posting numbers that reflect a moment of rare clarity in the semiconductor industry: the world needs what they produce, and there is nowhere else to go. The 210 trillion won figure is not just a quarterly forecast. It is a statement about where value concentrates in the age of artificial intelligence — at the bottom of the stack, in silicon wafers, in packaging houses, in factories that can run 24 hours a day and still fall short of demand.