SK Hynix Bets $38 Billion on AI Memory Boom with Two New Fabs

Inside SK Hynix's $38bn Yongin and Cheongju Fab Push

SK Hynix, the world’s second-largest memory chipmaker, has greenlit a colossal W54 trillion ($38.1bn) investment to construct two new fabrication plants in South Korea, betting that the artificial intelligence revolution will sustain voracious demand for advanced memory chips well into the next decade. The announcement underscores how the AI infrastructure buildout is reshaping capital allocation across the global semiconductor industry.

The larger of the two projects, the Y2 fab at the Yongin semiconductor cluster, will be dedicated to DRAM production, with a heavy focus on high-bandwidth memory (HBM) and other next-generation products. It carries a price tag of W35.2tn (around $24.8bn), with construction slated to begin in July 2027 and the first clean room expected to be operational by June 2029. The second facility, dubbed M17 in Cheongju, will focus on NAND flash memory, costing W19.1tn ($13.5bn) and starting construction in February 2027, with initial production targeted for December 2028.

The investment lands amid a punishing shortage of memory chips, fueled by soaring investments in AI infrastructure, data centers, and demand from chipmakers like Nvidia that rely on large quantities of HBM for their AI accelerators. Prices have surged in response, squeezing buyers across the electronics supply chain. SK Hynix itself acknowledged that technological leadership alone is no longer enough – "the ability to deliver the necessary quantities exactly when customers need them is the greatest competitive advantage."

However, the new fabs will not ease the near-term supply crunch. With Y2 not beginning production until mid-2029 and M17 in late 2028, the industry faces at least two more years of tight conditions. Counterpoint Research notes that Samsung, Micron, and China’s CXMT are also expanding, but so far demand is growing faster than the planned new capacity, meaning prices are unlikely to soften before late 2028.

How SK Hynix, Samsung and Micron Are Racing to Meet AI Memory Demand

SK Hynix’s Strategic Moonshot

By committing nearly $38bn to two greenfield fabs, SK Hynix is betting that the memory market’s AI-driven upcycle has fundamental, multi-year legs. The Y2 plant alone will substantially boost its HBM output – the most profitable and fastest-growing memory segment – where it already holds a leading position supplying Nvidia. The company’s own statement makes clear that it views on-time delivery, not just technology, as the decisive competitive weapon in the AI era.

The Supply-and-Demand Equation Until 2029

With M17’s first NAND wafers due only in December 2028 and Y2’s DRAM not until mid-2029, the capacity timeline effectively locks in the current shortage for at least two more years. Demand from AI servers, cloud data centers, and next-gen accelerator chips continues to outstrip supply, and even the combined expansions of all major players are unlikely to bring the market into balance before 2029. Consequently, memory contract prices are expected to remain elevated, sustaining unusually high margins for producers.

The Competitive Field: Samsung, Micron, and Chinese Challengers

SK Hynix is not acting in a vacuum. Samsung is concurrently expanding its Pyeongtaek campus, Micron has telegraphed capacity additions, and China’s CXMT is ramping up, although its ability to produce cutting-edge HBM at scale remains uncertain. The risk of oversupply looms if all these projects commence on schedule – a scenario that Counterpoint Research warns could materialise by 2028. Yet SK Hynix’s early mover advantage in HBM, coupled with its tight customer relationships, may allow it to defend market share and pricing even in a more crowded landscape.

What This Means for AI Chipmakers Like Nvidia

Nvidia and other AI accelerator designers depend on a stable, growing supply of HBM to boost the performance of their products. SK Hynix’s additional capacity, though years away, could eventually ease one of the most stubborn bottlenecks in the AI hardware chain. In the interim, AI firms and cloud providers will continue to face allocation challenges and high component costs, making long-term supply agreements increasingly critical.

For Chip Buyers and Investors: Navigating the Extended Memory Shortage

• Memory procurement teams: Expect DRAM and NAND contract prices to remain elevated through at least mid-2028. Lock in multi-year supply agreements now, particularly for HBM, before competitors pre-empt available allocation.
• Investors in SK Hynix: The $38bn capex program will depress near-term free cash flow and may lift leverage. Monitor quarterly results for signs of construction delays at Y2 or M17, which could extend the payback period and blunt the expected revenue windfall from the AI cycle.
• Competitors: Samsung and Micron have a narrow two-year window to capture HBM market share before Y2 comes online. Any slippage in their own ramp schedules would further tighten supply and strengthen SK Hynix’s pricing power.
• Data center operators and AI firms: Plan for high memory costs in infrastructure budgets through 2029. Diversify memory sources where possible, but recognize that cutting-edge HBM supply will remain concentrated among a handful of players, with SK Hynix firmly at the centre.