Inside the Nvidia-SK Hynix Alliance Reshaping AI Memory

A partnership beyond HBM4 supply

SK Hynix already supplies an estimated 50–70% of Nvidia’s HBM4 requirements, but the new agreement is designed to move the relationship from volume supply to joint roadmap planning . The two companies will co-develop next-generation memory solutions that are aligned with Nvidia’s next-generation GPU and system architectures, addressing the extended development cycles, advanced fabrication, and enormous capital investment required to sustain the global buildout of AI factories .

The platforms covered reveal how deeply memory is being integrated into Nvidia’s full-stack vision. The Vera Rubin platform represents Nvidia’s next AI supercomputing architecture, while RTX Spark targets the personal AI PC market and Jetson Thor powers autonomous machines and industrial robotics. By designing memory for all three simultaneously, Nvidia and SK Hynix are betting on a single memory roadmap scaled across cloud, client, and edge .

The structural memory shortage and the race to build capacity

Huang’s Seoul comments were the latest in a year-long chorus of warnings about a memory supply bottleneck that refuses to ease. At Computex 2026 in Taipei just days earlier, SK Group Chairman Chey Tae-won projected that supply-demand tightness would persist through 2030—a timeline that has been independently echoed by Samsung, which forecast “significant shortages” through at least 2027, and by Intel CEO Lip-Bu Tan, who warned the crunch could last until 2028 .

Chey made his most aggressive capacity commitment to date, announcing that SK Hynix would double its total wafer production capacity within five years . He noted that building a new memory fab requires “enormous investment and takes at least three years,” which explains why the supply picture is locked in for the rest of the decade . SK Group earlier stated at GTC 2026 that global memory chip supply would remain about 20% below demand through 2030 .

Bringing AI into chip design with PhysicsNeMo

One of the most technically consequential components of the broader Nvidia-SK relationship is the integration of AI into semiconductor manufacturing itself. SK Hynix is using Nvidia’s CUDA-X accelerated libraries, including the PhysicsNeMo framework, to apply AI-powered physics simulations that dramatically speed up technology computer-aided design (TCAD) simulations .

This initiative, first announced as part of the broader SK Group-Nvidia partnership at APEC in October 2025, is now moving into production use. The same October 2025 agreement also committed SK Group to building a major Nvidia AI factory in South Korea housing more than 50,000 GPUs, with the first phase expected by late 2027 .

Factory digital twins and the autonomous fab endgame

The most ambitious operational target in the partnership is SK Hynix’s plan to build completely autonomous fabs by 2030. The roadmap combines Nvidia Omniverse for 3D visualization and simulation, OpenUSD for scene optimization, and Nvidia cuOpt for logistics and scheduling .

SK Telecom, an SK Group affiliate, has already completed a proof-of-concept digital twin of an SK Hynix semiconductor fab, simulating production lines, logistics, and equipment with Nvidia Omniverse libraries . At GTC San Jose in March 2026, Nvidia named SK Telecom a key physical AI partner, and a formal joint committee was established to accelerate co-development of physical AI technology across semiconductor fabs, shipbuilding, and defense facilities .

The autonomous fab is being built around three core pillars: Operation AI, which acts as the brain; Physical AI, serving as the body; and the digital twin, which provides a safe environment to evolve both . SK Hynix plans to commercialize digital twin technology in stages in lockstep with the 2030 autonomous fab target .

Taiwanese supply chain ties and regional strategy

The Seoul and Computex meetings also revealed the broader regional supply chain dynamics at play. At Computex 2026, Chey Tae-won signaled stronger ties with TSMC and Taiwanese suppliers, while Nvidia has been increasing its spending commitments with Taiwanese supply chain partners to secure capacity across the full semiconductor stack . The deepening trilateral alignment between Nvidia, SK Hynix, and TSMC represents a strategic supply chain anchor for the next generation of AI infrastructure.

The multiyear deal between Nvidia and SK Hynix is not a standard memory procurement agreement—it is a structural bet on co-engineered hardware that will define AI factory economics for the rest of the decade. With memory projected to remain scarce through 2030, and new wafer capacity taking half a decade to bring online, the partnership locks both companies into a mutually dependent growth trajectory that spans silicon design, factory automation, and the physical AI economy.