NVIDIA CEO Jensen Huang has requested SK Hynix to accelerate the delivery of its next-generation HBM4 memory by six months. This move underscores NVIDIA's eagerness to advance its AI capabilities.
At the SK AI Summit in Seoul, Huang formally asked SK Hynix Chairman Chey Tae-won to expedite the HBM4 delivery, originally slated for the second half of 2025. This suggests NVIDIA aims to acquire the chips by early 2025.
While the exact reason remains undisclosed, it's likely a strategic move by NVIDIA to gain an early advantage in integrating HBM4 into its next-gen architectures, such as Rubin. Given HBM4's crucial role in future AI GPUs, NVIDIA is keen to avoid any design flaws that hampered its Blackwell architecture.
HBM4 represents a significant advancement in memory technology, integrating memory and logic semiconductors into a single package. This eliminates the need for separate packaging, improving performance and efficiency. This streamlined process also eases pressure on the CoWoS supply chain, involving companies like TSMC.
SK Hynix has reportedly taped out HBM4, a significant step towards mass production. While competitors like Samsung and Micron are also developing HBM4, SK Hynix currently holds the spotlight in this race.