SK hynix will deliver the first samples of HBM4 to Nvidia in June

The year has barely begun, and knowledgeable sources are already reporting on SK hynix’s further plans to expand the range of memory offered for computing accelerators. In June of this year, this South Korean company intends to begin supplying HBM4 samples for Nvidia’s needs, and by the end of the third quarter it is preparing to launch mass production of such memory.

Image Source: SK Hynix

As already noted, the commitments undertaken by SK hynix to accelerate the implementation of HBM4 by about six months allowed Nvidia to shift the announcement of the Rubin family of computing accelerators, which require such memory in a 12-tier version, from 2026 to the second half of the current year. Digital designs of HBM4 chips from SK hynix specialists were ready by the end of the last quarter, and now they are beginning to closely interact with colleagues from Nvidia on the issue of introducing HBM4 into mass production. Each Rubin accelerator will require eight stacks of HBM4, and the later Rubin Ultra will use 12 stacks of HBM4.

Competing Samsung Electronics will try to complete the development of HBM4 by the middle of this year, and expects to produce DRAM chips for the corresponding stacks using more advanced 10-nm class technology than SK hynix and Micron are planning to do. The latter intends to launch production of HBM4 only in 2026. In the case of the release of HBM3E for Nvidia’s needs, SK hynix remains the market leader, but competitors will try to get ahead of it in at least some way in order to receive profitable orders from the largest developer of computing accelerators.