In Seoul, the trading day began with a headline and ended with a denial. SK Hynix shares fell as much as 4 percent in early trading on reports that the company was in talks to acquire Intel’s wafer plant in Ohio, and the stock recovered only after the company issued a statement: it is not negotiating to buy the Ohio fab. By the close, the shares had erased most of the morning’s losses, and the company had moved on to the news it actually wanted to make.
The announcement that followed was the one that fits its strategy. SK Hynix said it will invest 7.09 trillion won, about $5 billion, in its Cheongju AI memory complex in South Korea, with the money going mainly to expand advanced packaging capacity for high-bandwidth memory, the stacked chips that sit beside every AI accelerator and that SK Hynix dominates. The two headlines, read together, amount to a sentence the company has been writing for two years: it does not want old logic capacity, it wants AI memory, and it will spend wherever that edge is.
The Ohio report was always a stretch on the merits, analysts said. Intel’s campus in New Albany, Ohio, was conceived as a giant of U.S. chip manufacturing, but Intel delayed its completion last year, with production now not expected until around 2030, and the project’s future has been a subject of speculation as Intel restructures its foundry business. A memory maker buying a logic fab made little sense: the processes, the customers and the economics are all different, and SK Hynix has no experience running the kind of leading-edge logic manufacturing that Intel’s Ohio site was designed for.
The denial put the speculation to rest quickly, and the market’s reaction showed how much of the morning’s drop was noise. What matters for SK Hynix is the memory market, where the dynamics could hardly be stronger. The company is the leading supplier of high-bandwidth memory to Nvidia, holding roughly half of the global market, and it has sold out its capacity through the current generation of accelerators. Every new AI data center built this year needs HBM stacks, and the constraint in the chain has moved from the accelerator to the memory that surrounds it.
The Cheongju investment extends a buildout that was already among the largest in the industry. The company is constructing a new advanced packaging plant, P&T7, at the site, a 19 trillion won project, about $13 billion, that broke ground in April and is scheduled for completion by the end of 2027. The new 7.09 trillion won commitment adds packaging capacity for the next generation of HBM, which requires more complex stacking, more precise bonding and more testing than the current products, and which commands higher prices.
Cheongju has become SK Hynix’s AI memory cluster. The complex already hosts the M11, M12 and M15 fabs, the M15X facility built for next-generation DRAM and HBM, and the P&T3 packaging and testing plant. The company said the new investment will expand capacity across the site, taking advantage of the proximity of wafer production and packaging in one location, which shortens cycle times and cuts logistics costs. The economics of memory are brutal and cyclical, but the current cycle has been anything but typical: demand from AI has absorbed every unit the industry can produce.
The broader market is investing on the same logic. Samsung, SK Hynix’s crosstown rival, is spending heavily on its own HBM capacity and has won a larger share of the latest generation of orders. Micron, the third major supplier, has raised its own capital plans and expanded its U.S. operations. All three are betting that AI demand for memory lasts long enough to justify capacity that, in an earlier cycle, would have been considered reckless. The risk is the same one that has always haunted memory: a sudden drop in demand leaves the industry holding factories it does not need.
SK Hynix’s answer to that risk is the same as its answer to the Ohio rumor: focus on what it does best. The company’s technology, including the packaging process it uses to stack memory chips, gives it a cost and performance edge that rivals have been slow to match, and its position as Nvidia’s primary memory supplier has proved durable through multiple product generations. The $5 billion commitment is a bet that edge holds.
The day’s two stories, the denial and the investment, were really one story about capital allocation. SK Hynix could have chased a dramatic acquisition in American logic manufacturing, the kind of deal that makes headlines and complicates balance sheets. Instead it said no to the fab and yes to the memory, a choice that tells investors exactly where the company believes the value is: in the stacks of silicon that sit next to the world’s most expensive chips.
