Micron's HBM4 Capacity Ramp Accelerates by Double, mass production of HBM4E by 2027

Korean media outlet THE ELEC reported on Monday that Micron Technology's production ramp-up of its sixth-generation high bandwidth memory HBM4 is progressing smoothly, with the speed being twice as fast as last year's HBM3 12-layer product and concurrent yield improvements.

Micron's Vice President of Global Operations, Manish Bhatia, disclosed at the JPMorgan Chase investor conference that HBM4 will be equipped on NVIDIA's next-generation AI computing platform "Vera Rubin," thereby establishing Micron's position as a key supplier.

The acceleration behind this ramp-up is supported by three pillars: First, the rich experience and learning effects accumulated from the mass production of HBM3 and HBM3E; Second, the core Die uses the mature 1β (10nm 5th generation) process, whose performance and yield have been fully verified; Third, by pairing with self-developed basic Dies, the product completion and performance are maximized.

In the more advanced HBM4E generation, Micron will initiate a significant strategic shift. The core Die will be upgraded to the 1γ (10nm 6th generation) process that introduces EUV lithography for the first time, which is a contemporary technology to Samsung and SK Hynix's sixth-generation 1c process; the basic Dies will no longer be manufactured in-house but will be outsourced to TSMC for manufacturing. Micron plans to mass-produce in 2027, with the first batch launching JEDEC standard products and simultaneously developing customized versions.

Samsung and SK Hynix are also advancing their HBM4E deployments at a tight pace. Samsung is expected to send out the first batch of samples in the second quarter of this year, with the basic Dies still being manufactured in-house; SK Hynix plans to send samples in the second half of the year and mass-produce in 2027, with the basic Dies also being produced by TSMC, which is reported to adopt a 3nm process.

In terms of the process iteration rhythm, Micron has set its targets in sync: by mid-2026, the shipments of 1γ DRAM and the ninth-generation NAND will account for more than half of the total capacity, at which point 1γ will become the company's largest single DRAM process node.

Industry insiders point out that with the surge in AI compute power demand, the HBM technology race has entered a white-hot stage. With its deep collaboration with TSMC, Micron is expected to further narrow the gap with Korean giants in high performance and yield control, accelerating the competition for dominance in the AI server memory market.