Recent revelations from SK hynix management made it clear that the company allegedly accelerated the development of HBM4 on its own initiative, but South Korean media today explained that Nvidia at least did not object to this, since it is going to introduce Rubin accelerators with this type of memory in the third quarter. Samsung has also accelerated its efforts and is now preparing to finalize the HBM4 for mass production in the current half of the year.
Image Source: Samsung Electronics
It was originally planned that Samsung would finally begin shipping HBM3E chips to Nvidia in the first half of this year, followed by HBM4 in the second half of the year, but now the latter stage is planned to be accelerated by almost six months. At the very least, this will allow Nvidia, if necessary, to provide a second source of HBM4 memory supplies by the time the Rubin family of accelerators are announced in the third quarter. Nvidia originally expected to introduce Rubin in 2026, but now with access to memory like HBM4 it could speed up.
As noted earlier, one of the features of HBM4 memory is the presence of a base crystal, whose functional features are customized to the needs of a specific customer. Samsung expects to produce such crystals for HBM4 on its own using 4nm technology, while SK hynix will depend on TSMC in this area. However, this gives SK hynix the opportunity to immediately receive 3-nm base crystals from TSMC, while Samsung will depend on the success of its contract division in mastering advanced technological processes.