Samsung is reportedly planning to more than double production of its HBM4 and HBM4E memory chips, feeding surging AI accelerator demand.
Samsung is expected to more than double output of its next-generation HBM4 and HBM4E DRAM in the coming year, according to sources cited by South Korean media. The chips are critical high-bandwidth memory components used in AI training and inference accelerators from Nvidia and other GPU makers.
The scale-up reflects continued tightness in the memory supply chain as AI infrastructure buildouts accelerate globally, with memory increasingly the binding constraint on GPU system availability rather than compute logic itself.
Memory supply, not just GPU allocation, is becoming the real bottleneck for AI infrastructure buyers — capacity planning and vendor negotiations need to account for HBM lead times now. Expect this to ease some 2027 pricing pressure on AI training clusters if Samsung executes.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →