The Memory King

From Hyundai's Garage to the Centre of the AI Universe

The story of SK Hynix begins in 1983, when Hyundai Group founder Chung Ju-yung decided South Korea needed its own semiconductor industry. Hyundai Electronics was born in a country that had been rebuilding from war for barely three decades, with the explicit ambition of competing with Intel and Japan's NEC. It was an audacious bet. The company survived the Asian financial crisis of 1997, a near-bankruptcy in 2001, a merger with LG Semiconductor, and a $3 billion acquisition by the SK Group conglomerate in 2012 — the moment it became SK Hynix. Through all of it, the engineers kept pushing the physics.

Today, SK Hynix is the world's second-largest DRAM manufacturer and the world's dominant supplier of the specific type of memory that AI needs most. That distinction — between ordinary DRAM and HBM — is the entire thesis.

What Is HBM, and Why Does It Matter So Much?

A standard DRAM chip sits beside a processor and communicates with it through a relatively narrow channel. It is fast enough for a laptop, a phone, a database server. It is not fast enough for a large language model processing billions of parameters simultaneously. The bottleneck is not compute — it is memory bandwidth. The AI chip cannot think faster than it can read.

HBM solves this by stacking multiple DRAM dies vertically, connecting them through thousands of microscopic through-silicon vias, and mounting the entire stack directly on the same substrate as the processor using a technology called 2.5D packaging. The result is a memory chip that delivers bandwidth of up to 1.2 terabytes per second — roughly ten times what conventional DRAM can achieve — in a package small enough to sit beside an NVIDIA H100 or B200 GPU.

Every NVIDIA AI accelerator ships with HBM. Every Google TPU ships with HBM. Every Microsoft Maia 200, confirmed in January 2026 to use 216GB of HBM3E exclusively supplied by SK Hynix, ships with HBM. The AI infrastructure buildout — the $300 billion annual capex cycle that the hyperscalers have committed to — is, at its physical core, a massive order for SK Hynix's most advanced product.

The Numbers Behind the Dominance

SK Hynix's position in the HBM market is not a narrow lead. It is a structural advantage built over years of early investment that its competitors are still trying to close.

Samsung's HBM3E yield problems — widely reported throughout 2024 and into 2025 — handed SK Hynix a window that it used to deepen its relationship with NVIDIA, Google, and Microsoft. By the time Samsung's quality issues were resolved, SK Hynix had locked in long-term supply agreements and begun mass production of HBM4, the next generation, ahead of schedule.

In September 2025, SK Hynix became the first company in the world to begin mass production of HBM4. In March 2026, NVIDIA confirmed that SK Hynix and Samsung — not Micron — would be the exclusive suppliers of HBM4 for the Vera Rubin platform, with SK Hynix receiving approximately 70 percent of the allocation. Micron, despite aggressive investment, was not invited.

The Supercycle Thesis

The memory industry has historically been cyclical — boom and bust, driven by oversupply. The bear case for SK Hynix is that this cycle will end as all cycles do, with prices collapsing when supply catches up to demand. It is a reasonable concern, and one that the company's own analysts acknowledge.

But there is a credible argument that this cycle is structurally different, and it rests on three pillars.

The Infrastructure Behind the Lead

What makes SK Hynix's position genuinely difficult to replicate is not just the technology — it is the infrastructure being built around it.

The Risks Are Real

No investment thesis is complete without an honest account of what could go wrong, and SK Hynix carries genuine risks.