SK Hynix Sees A Historic Memory Crunch Coming
er of 2028. That creates a tightrope for SK Hynix: it’s expanding capacity in South Korea and investing in the US, but new factories and advanced packaging lines have long lead times, and a sudden wave of new supply can quickly flip a shortage into a glut. For now, though, big cloud companies keep spending heavily on data centers, which suggests demand for AI hardware is still running ahead of what the memory supply chain can comfortably deliver.
Why should I care?
For markets: UBS sees DRAM undersupply lasting to at least the second quarter of 2028.
When memory stays scarce, pricing is set less by “spot” swings and more by who gets guaranteed supply in long-term contracts. That tends to give leading suppliers like SK Hynix more pricing power, but it also makes profits unusually sensitive: a small change in expected supply additions, or in AI-server demand, can have an outsized effect on selling prices. The same bottleneck can also raise the all-in cost of AI computing for customers buying Nvidia-based systems, which means this isn’t just a chipmaker story: it can influence how quickly companies can roll out new AI capacity.
