Samsung and SK Hynix make over 80% of the world's memory chips and are racing to supply Nvidia and OpenAI, even as US export controls tighten around their older plants in China.
However expensive Nvidia's GPUs get, they still need a component called high-bandwidth memory to swallow data fast enough to train and run AI models. More than 80% of that memory comes from two Korean companies: Samsung and SK Hynix.
Chasing Nvidia
SK Hynix bet on the right technology early and has long been Nvidia's main supplier. Samsung is catching up: its market share more than doubled over the past year, and it has signed deals with both Nvidia and OpenAI.
Samsung is building a packaging plant in Indiana; SK Hynix is spending $4 billion on a US facility of its own. What they are selling is not just chips but their own indispensability.
Who decides
As one analyst puts it, memory market share still comes down to how Nvidia allocates orders. Whether Samsung closes the gap depends on how much of Nvidia's next-generation chip supply it can take from SK Hynix.
Caught in the middle
Both companies have legacy plants in China, which accounts for roughly a third of the global chip market. But in 2025 Washington scrapped the licence-free export channel and replaced it with annual reviews, while domestic Chinese suppliers are growing fast.
Take America's business and you anger China's customers; stay in China and you risk losing the global lead.
Why it matters
AI compute is described as the new oil, but the pipeline is not held by Nvidia — it is held by two Korean memory makers. Their capacity and their allegiances decide how fast and how cheaply the next generation of AI infrastructure gets built, and who gets pushed off the bus in the US-China chip war.
Curated from high-quality sources, with concise summaries and key takeaways.
