SK hynix approves $38B+ investment in two new memory fabs
SiliconANGLE Maria Deutscher
SK hynix approved $38.3B for two new memory fabs. One will make HBM for AI chips; the other adds NAND flash capacity.
Based on reporting by SiliconANGLE, Maria Deutscher — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
SK hynix’s board has signed off on a huge buildout: two new memory fabs that will cost 54 trillion South Korean won, or about $38.3 billion. The project sits inside a broader $430 billion plan to raise manufacturing capacity, and it also ties into South Korea’s wider push to grow its chip industry.
The bigger-ticket site is Yongin Y2, a $25 billion fab at the Yongin Semiconductor Cluster, about 20 miles south of Seoul. SK hynix says it will be the second of four chip factories planned there. The first fab in the cluster is due to open next February, while construction on Yongin Y2 is set to begin in July 2027. Its first cleanroom is scheduled for June 2029, and the finished facility will cover 279 acres.
Yongin Y2 will be devoted entirely to DRAM. That includes HBM, the fast memory used by graphics cards to hold AI models and the data they work on. SK hynix is already the world’s largest HBM producer, and last September it introduced its most advanced version so far: a 12-layer chip based on HBM4.
The second new fab will focus on NAND flash. It will go up on a campus that already has three SK hynix factories, which should help because the power and water hookups are already there. SK hynix expects the first cleanroom at that site in December 2028. The company and Sandisk also unveiled a new NAND approach earlier this year, called HBF, that stacks NAND flash circuits in a way similar to HBM and is said to match HBM’s performance while offering up to 16 times more capacity.
Samsung is pushing in the same direction. On Wednesday, SK hynix’s biggest rival showed off two technologies that let NAND and HBM be stacked directly on top of logic circuits, cutting the distance data has to travel. Samsung says its stackable HBM setup could be eight times faster than HBM5, which is expected near the end of the decade, and could raise memory density by ten times. The memory race is getting very vertical.
My take — AI-written commentary, not fact-checked reporting
This is what real competition looks like: not louder hype, just more expensive factories and more layers of silicon. The AI boom has turned memory into the bottleneck everyone politely ignored until the bill arrived. Samsung and SK hynix are now racing to stack their way around physics, which is a very modern kind of stubbornness.
Read more about this at: SiliconANGLE