Nvidia and SK Hynix sign multi-year pact to develop next-gen AI memory chips

Nvidia and SK Hynix have locked in a multi-year agreement to co-develop next-generation memory chips purpose-built for AI workloads. The deal, announced on June 7 following a meeting between Nvidia CEO Jensen Huang and SK Group Chairman Chey Tae-won in Seoul, centers on HBM4, the latest iteration of high-bandwidth memory that will power Nvidia’s upcoming Vera Rubin accelerator platform.

This isn’t just a supply agreement. It’s a full design-and-manufacturing collaboration that spans chip architecture, infrastructure buildout, and product diversification across personal AI and robotics applications. Initial deliveries are slated for Q3 2026.

What the deal actually covers

The partnership positions SK Hynix as Nvidia’s largest memory partner. The collaboration covers components for several Nvidia product lines: the Vera Rubin accelerator, Vera CPUs, RTX Spark PCs, and Jetson Thor robotics systems, spanning data center AI, consumer computing, and physical AI.

SK Hynix has also signaled plans to double its wafer production capacity over the next five years.

Why memory is the real AI bottleneck

High-bandwidth memory solves bandwidth constraints by stacking memory dies vertically and connecting them with ultra-fast interconnects, dramatically increasing data throughput compared to traditional DRAM. SK Hynix pioneered HBM technology back in 2013, and each subsequent generation has delivered meaningful jumps in bandwidth and energy efficiency.

The two companies have history here. SK Hynix supplied HBM3 chips for Nvidia’s H100 GPU, the workhorse accelerator that powered much of the initial generative AI boom. That earlier collaboration served as the proving ground for this deeper, longer-term partnership.

The competitive landscape and what it means for investors

SK Hynix, Samsung, and Micron have all been approved as HBM4 suppliers, but the depth of this partnership gives SK Hynix a meaningful head start on integration with Nvidia’s platforms. Samsung, in particular, has been playing catch-up in the HBM segment after quality issues delayed its entry into supply for Nvidia’s latest GPU generations.

SK Hynix’s plan to double wafer capacity over five years is ambitious, but even that timeline suggests supply constraints won’t ease quickly.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.