Analysis
The Buildout
SK Hynix is spending an estimated $720 billion to triple its memory-chip production capacity by 2034, according to CNBC, which secured the first on-camera tour of the company's new Yongin Cluster campus south of Seoul and additional cleanroom access at its existing Cheongju site. The first fab at Yongin will reach the height of a 50-story apartment building once complete -- a physical scale that mirrors the size of the financial commitment behind it.
Why Memory, Specifically
SK Hynix makes the majority of the world's high-bandwidth memory (HBM) chips, the specialized memory that sits directly alongside GPUs in AI accelerators and has become one of the tightest bottlenecks in AI hardware supply. Unlike logic chips, where TSMC dominates manufacturing, HBM supply is concentrated among SK Hynix, Samsung and Micron -- and SK Hynix has taken the clearest lead, which is why Nvidia secured a stable HBM supply commitment and agreed to co-develop next-generation memory as part of a separate $500 billion deal with SK Group that also includes new data centers built with SK Telecom by 2027. That Nvidia deal came on top of an SK Hynix announcement on August 7 to spend $38 billion building new memory chip plants specifically to meet soaring demand.
The Historical Risk
Memory chips have one of the most punishing boom-and-bust cycles in semiconductors -- oversupply has repeatedly crushed prices within a year or two of a capacity buildout, and SK Hynix itself went through a sharp downturn as recently as 2023. Committing $720 billion toward tripled capacity is a bet that AI-driven HBM demand structurally breaks that cycle rather than repeating it, a wager the company's SEC filings and SK Group Chairman Chey Tae-won's own comments to CNBC suggest leadership is making with open eyes about the historical pattern, not despite it.
Company Background
SK Hynix debuted on U.S. markets via a dual listing covered by CNBC in July, crossing a trillion-dollar valuation and giving American investors direct exposure to a company whose fortunes were previously accessible mostly through Korean exchanges. The stock has been volatile in recent weeks -- it was part of a sharp Kospi-wide selloff in late July before rebounding -- reflecting how tightly SK Hynix's valuation now tracks sentiment about AI infrastructure spending broadly, not just its own execution. Pulse previously covered that Nasdaq debut.
The Competitive Field
Samsung and Micron are both racing to expand their own HBM capacity, but SK Hynix's early technical lead and its direct supply relationship with Nvidia give it first-mover advantage in the highest-margin segment of memory. Samsung has historically outspent SK Hynix on total semiconductor capex, so the two companies' buildout plans will determine whether HBM supply loosens meaningfully before 2034 or stays as constrained as it has been through 2026.
The Counterweight
A 2034 target is eight years out -- an unusually long horizon for a capital commitment this size in an industry where AI hardware architectures have shifted meaningfully every year or two. If a future generation of AI chips needs meaningfully less memory per unit of compute, or if a rival memory technology displaces HBM, SK Hynix's tripled capacity could arrive into a market that no longer needs it at the price the company is underwriting today. The $720 billion figure also isn't fully committed capital -- it's a multi-year plan subject to revision if AI capex growth slows industry-wide the way it has in prior cycles.
