SK hynix bets $38 billion on new fabs to feed AI memory demand

SK hynix is making another massive wager on artificial intelligence. The South Korean memory giant approved 54 trillion won (about $38 billion) in new manufacturing investments on Friday, authorizing two new chip fabs that will expand production of the memory chips powering AI systems. The move reflects the company’s belief that demand for AI infrastructure will stay strong for years—and that memory, not processors, remains the industry’s biggest constraint.

This announcement comes in the wake of record quarterly results that have demonstrated the importance of high-bandwidth memory (HBM) in the current AI boom. As per a previous report by Cryptopolitan, SK Hynix posted revenue of 79.3 trillion won in the second quarter, and 60.5 trillion won in net income due to growing demand for HBM technology. The company has begun to make large deliveries of HBM4 chips in order to maintain its leadership in the field of AI memory.

The investment is crucial for firms that manufacture AI chips and cloud services because sophisticated accelerator devices rely on multiple HBM stacks being integrated next to chips using advanced packaging technology, such as TSMC’s CoWoS. Merely increasing GPU chip production would not help if HBM memory is not available in enough numbers.

According to a report published by Omdia, memory is likely to account for over 50% of the global semiconductor market revenue by 2026 due to the increased demand for DRAM and NAND on account of AI. TrendForce claims that the supply of DRAM and HBM will remain tight at least through 2027, as the capacity for advanced packaging will be limited despite several industry-wide developments.

SK hynix provides HBM to Nvidia and other manufacturers of AI accelerators, meaning its manufacturing intentions become a key measure of the speed of expansion of AI infrastructure. This investment follows Cryptopolitan’s recent report about Nvidia’s CEO Jensen Huang’s announcement of $500 billion cooperation with SK Group, demonstrating the increasing significance of memory production.

Where the 54 trillion won goes

SK hynix announced that a sum of 35.2 trillion won will be allocated for the construction of its Y2 DRAM fab located in Yongin. At the same time, 19.1 trillion won will be used for building the M17 NAND flash fab located at its Cheongju campus. It is estimated that the combined investment will amount to around $38 billion.

These projects are an integral component of SK hynix’s more expansive plans for development. The company has already committed 600 trillion won to the Yongin Semiconductor Cluster and another 100 trillion won to expand Cheongju, while construction of the neighboring Y1 fab is already underway.

The firm is taking steps to fortify its position in AI memory. SK hynix received 58% of global HBM revenue in the initial quarter of 2026 based on Counterpoint Research’s Memory Tracker, with Samsung (21%) and Micron (21%) following behind.

HBM is now one of the rare materials in the AI industry. Different from regular DRAM materials, it makes use of stacking technology, through-silicon vias technology (TSV), and complex packaging technology, making it extremely hard to fabricate.

As per Omdia’s projections, there will be a 94.1% growth in the semiconductor industry in the year 2026, spurred by AI-related demand for NAND and DRAM.

The shortage of supply has already started to impact plans for product development. For instance, Nvidia is now considering 8-Hi HBM and HBM4 solutions for its next Rubin Ultra accelerator instead of the previously planned 12-Hi HBM4e due to worries that reduced availability of DRAM in 2027 may hinder production of HBM modules.

This dynamic means that suppliers of memory increasingly control how quickly AI hardware can enter the market. An increase in memory capacity leads to the growth in shipments of accelerators, while a shortage of memory forces manufacturers to revise the design of their products to the amount of memory available.

What the plants will build, and when

Analysts anticipate continued supply-tightening conditions because it will be several years before the new fabs are operational.

According to SK hynix, the Y2 fab is expected to produce both HBM products as well as next-generation DRAM products. Construction will start in July 2027, with the cleanroom expected by June 2029. The breaking ground for the M17 fab will take place in February 2027, and its cleanroom will be ready by December 2028.

According to SK hynix, this investment serves as groundwork for a transition that is expected to take time instead of just being a short-lived AI trend. The firm indicates that based on Omdia forecasts, it will see a 19 percent CAGR growth in the global demand for DRAM and NAND till 2030, thus confirming SK hynix’s belief in the relevance of mass memory manufacturing for competitively winning in the age of artificial intelligence.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.