Micron Technology’s latest high-bandwidth memory chip is scaling production at roughly double the speed of the generation before it.
The company’s HBM4, its sixth-generation high-bandwidth memory product, began volume shipments in March 2026 for Nvidia’s Vera Rubin AI platform. Micron has already fully allocated its entire calendar year 2026 HBM supply, including HBM4.
The numbers behind the ramp
Manish Bhatia, Micron’s Executive Vice President of Global Operations, laid out the HBM4 progress during a J.P. Morgan investor conference on May 20, 2026. The core claim: HBM4 is ramping at approximately twice the speed of the 12-high HBM3E generation, with yields improving faster than expected.
Bhatia attributed the acceleration to three factors: operational learnings carried forward from earlier production cycles, design simplifications baked into the HBM4 architecture, and supply chain optimizations that reduced bottlenecks.
On the performance side, HBM4 delivers speeds exceeding 11 Gbps per pin and achieves over 2 terabytes per second bandwidth per stack. Each stack holds 36 GB of capacity. That’s more than double the bandwidth of HBM3E in 12-high configurations.
Fiscal results that surprised everyone
Micron’s fiscal Q3 2026 results, released on June 24, 2026, provided concrete financial evidence that the HBM business is delivering. The company posted adjusted earnings per share of $25.11, compared to the Wall Street consensus estimate of $20.49. Revenue came in at $41.46B against expectations of $35.69B.
Micron exceeded the EPS estimate by roughly 23% and the revenue estimate by about 16%. Micron noted that it has surpassed $1B in HBM revenue. Micron has historically trailed SK Hynix in the HBM space. SK Hynix was the first to supply HBM3E to Nvidia and has enjoyed a dominant position as the preferred memory supplier for AI accelerators.