Micron is pushing harder into HBM as AI demand turns high-bandwidth memory into one of the industry’s more useful growth engines.
According to ETNews the US memory maker is expanding HBM capacity, with output expected to rise by as much as 60,000 wafers per month. Its industry deep throats put Micron’s HBM capacity at about 40,000 to 50,000 wafers per month last year, before the latest expansion push.
One source expects Micron’s total HBM capacity to reach roughly 100,000 wafers per month by the end of 2026. The shift is being driven by stronger demand for 12-Hi HBM4, which is used in Nvidia’s latest Vera Rubin AI accelerator platform.
Micron’s HBM output is still weighted towards 12-Hi HBM3E, but 12-Hi HBM4 is expected to take a much larger share this year. ETNews estimates 12-Hi HBM4 made up 20 to 30 per cent of Micron’s output early this year, potentially reaching 50 per cent by year-end.
Micron has said HBM4 production is scaling twice as fast as HBM3, helped by AI demand that continues to stretch memory supply. The company expects to start ramping HBM4E in 2027, with early samples using DRAM modules built on its 1-gamma process node.
Micron remains smaller in HBM than its Korean rivals, with Samsung and SK hynix each estimated at 150,000 to 200,000 wafers per month. Samsung is trying to close the gap with SK hynix by dedicating around half of its HBM capacity to HBM4 production.
Samsung Electronics is reportedly processing about 150,000 DRAM wafers per month for HBM, with roughly 75,000 going into HBM4. The company’s HBM4 yield has reportedly climbed to nearly 80 per cent, up from about 60 per cent six months earlier.
Nvidia is expected to diversify its HBM supply and configurations, including more 8-Hi HBM4 shipments from Samsung and SK hynix later this year.
