The Supply Picture: SK Hynix, Samsung, Micron, and a Constrained Market
Three companies make essentially all the high-bandwidth memory in the world: SK Hynix, Samsung, and Micron. That concentration, layered on top of a manufacturing process with painfully low yields and long lead times, makes HBM the tightest link in the AI supply chain. My read: memory is sold out generations in advance, qualification into a flagship accelerator is worth more than raw capacity, and the balance of power among the three makers is set by who qualifies first into the next GPU. Supply is not a footnote to the bandwidth story. It is the story's rate limiter.
Knowledge check
Why does qualifying into a leading accelerator's memory slot matter more to an HBM maker than simply having large manufacturing capacity?
Show the answer
HBM must pass stringent, accelerator-specific qualification before it can ship in a flagship GPU, and the vendor buys the whole generation's supply from whoever qualifies first and best. Capacity that is not qualified into the leading part cannot capture the high-margin demand, so qualification, not raw wafer volume, determines which maker wins the generation.