SK Hynix builds none of the models. It trains no agents. It has no consumer brand in the West. Yet without it, GPT-5 doesn’t run, Gemini doesn’t render, and every AI data center on the planet goes dark. On June 25, Bloomberg reported that the Korean chipmaker is seeking roughly $29.4 billion in a US stock listing — one of the largest IPOs in history. This is not a tech debut. It is a geopolitical signal.

The Company That Runs Under Every AI Model

There is a short list of companies that are genuinely irreplaceable in the AI supply chain. NVIDIA designs the GPUs. TSMC fabricates the chips. And SK Hynix manufactures the high-bandwidth memory — the HBM stacks — that attach directly to those GPUs and feed them data fast enough to run trillion-parameter models at scale. Without HBM, an H100 is a paperweight. Without SK Hynix, there is essentially no HBM market.

That is not an exaggeration. SK Hynix holds the dominant share of HBM3 and HBM3E production — the generations currently powering NVIDIA’s Hopper and Blackwell GPU families. Samsung is the only meaningful competitor on the supply side, and its HBM yield rates have lagged significantly behind SK Hynix’s. Micron is the third player in the broader DRAM market but has been even further behind on HBM ramp. For the AI data center build-out happening right now — the one being funded by a $94.5 billion HBM supercycle — SK Hynix is the critical path.

This is what makes a $29.4 billion US listing something more than a capital markets event. It is the moment the company that quietly enables every frontier AI model steps into the spotlight of Western finance.

Why List in the US, Not Seoul?

SK Hynix is already listed on the Korea Stock Exchange. Its market cap there is substantial. The decision to pursue a separate US listing — at the scale Bloomberg reports — is a deliberate strategic choice, not a routine fundraising exercise.

Three forces are converging to make this move logical now. First, US capital markets give SK Hynix access to the deepest pool of technology investors in the world — fund managers who understand semiconductor cycles, AI infrastructure economics, and long-duration capex bets. Second, a US listing improves SK Hynix’s standing in the ongoing US-China chip war. Washington has made clear it wants critical AI infrastructure inside its economic orbit. A Korean chipmaker listed on a US exchange, with a US shareholder base, is a different geopolitical actor than one whose capital is purely Asian. Third, the company needs the capital. Building out HBM4 and HBM4E manufacturing capacity — the next two generations that will be required for NVIDIA’s 2027-2028 GPU roadmap — is extraordinarily expensive. A $29.4 billion raise funds that buildout without stretching the balance sheet.

The timing is also pointed. The US CHIPS Act is actively paying semiconductor companies to build or expand on American soil. South Korea’s own K-Chips Act mirrors it. The political incentive structure around semiconductor supply chains has never been more favorable to this kind of cross-border capital move.

The Micron Comparison: Two Different Plays for Western Capital

The week of SK Hynix’s IPO disclosure is also the week that Micron — SK Hynix’s closest US-based competitor — deepened its partnership with Anthropic. The Micron-Anthropic four-pillar deal, which includes a Series H investment from Micron, signals that the memory oligopoly players are each seeking anchor relationships with frontier AI labs. Micron is buying proximity to Anthropic’s roadmap. SK Hynix’s existing relationship with NVIDIA remains its primary channel.

The two moves, read together, reveal the strategic logic: memory companies need locked-in demand visibility to justify the $15-20 billion capex cycles required to build new HBM generations. An equity relationship with an AI lab — or in SK Hynix’s case, a public US listing that puts it inside the portfolios of the same funds that own NVIDIA and Microsoft — is a way of securing that demand signal structurally, not just contractually.

Micron can already access US capital freely. SK Hynix is choosing to join the table. That choice says something about where the company believes the AI infrastructure investment cycle is heading: not a Korean story anymore, but a global one centered on American capital allocation.

The $35.5 Billion Operating Income Thesis

SK Hynix has projected $35.5 billion in operating income by 2029. To contextualize that number: it would make SK Hynix one of the most profitable semiconductor companies in the world — in the same neighborhood as TSMC’s current operating income, and exceeding what NVIDIA itself was generating two years before its recent earnings surge.

The projection rests on a single structural assumption: AI capex does not slow. The current generation of GPU clusters requires HBM at roughly 1:8 memory-to-compute ratio. Next-generation architectures — those required for trillion-parameter multimodal models at inference scale — are expected to require significantly more memory bandwidth, not less. Each successive GPU generation has increased HBM content per die. The roadmap strongly favors the memory suppliers.

The risk is concentration. SK Hynix’s HBM business is disproportionately dependent on NVIDIA. If NVIDIA’s GPU demand softens — whether from macro conditions, model efficiency gains that reduce compute requirements, or competition from custom silicon — SK Hynix’s revenue projections compress rapidly. A US listing allows it to diversify that story to investors who can model the full range of outcomes, not just the bull case. But it also means the story will be stress-tested publicly, by analysts and short sellers, for the first time.

The Strategic Read: What This Means for the AI Supply Chain

Three things are worth tracking as this listing advances:

1. The memory oligopoly is moving to secure its position before the cycle peaks. SK Hynix’s 2029 operating income projection implies continued exponential AI capex. A $29.4 billion raise now — before the peak, not after — signals management believes the next three years require significant additional capital investment to maintain the HBM production lead. Companies don’t raise $29 billion because they feel comfortable. They raise it because they see the competition closing.

2. This is a test of whether Western investors understand the AI infrastructure stack. Most retail investors who own NVIDIA understand what it does. Most do not have equivalent clarity on why HBM is as foundational as the GPU itself — or that without SK Hynix’s continued investment, NVIDIA’s roadmap stalls. The IPO process will force that education. When it succeeds, it changes how the market values the entire memory supply chain.

3. The geopolitical dimension is the real story. This is not a Korean company seeking Western prestige. It is a critical piece of global AI infrastructure choosing to anchor itself in US capital markets — with all the alignment of interest that implies. At a moment when the US government is spending hundreds of billions to bring semiconductor supply chains onshore or into allied orbits, a voluntary $29.4 billion listing is the most efficient sovereign alignment tool imaginable. No mandate required. The market does it.

FourWeekMBA · AI News · Published June 25, 2026 · Source: Bloomberg