NVIDIA and SK hynix sign multiyear memory partnership for AI factories
NVIDIA and SK hynix have announced a multiyear technology partnership to co-develop advanced memory aligned to NVIDIA's AI infrastructure roadmap, spanning data-centre AI supercomputers, personal computing and robotics. The agreement, announced on 7 June 2026, formalises what both companies describe as years of deep co-engineering and is structured to address the extended lead times and capital intensity required to keep memory supply in step with accelerating AI factory deployments.
Under the partnership, SK hynix will develop memory for four NVIDIA platform families: the Vera Rubin AI supercomputer, the Vera CPU, RTX Spark-powered personal AI PCs and the Jetson Thor robotic computing module. No financial terms, revenue commitments or volume figures were disclosed in the announcement.
AI applied to chip design and manufacturing
Beyond supply, the two companies are applying AI tooling to semiconductor design and fabrication. SK hynix is using NVIDIA's CUDA-X libraries and the PhysicsNeMo physics-simulation framework to accelerate technology computer-aided design (TCAD) workflows, computational lithography and in-house simulation codes — work NVIDIA says could extend to three-way collaborations involving electronic design automation software vendors.
Jensen Huang, founder and chief executive of NVIDIA, said: "Together, we will codevelop the next generation of memory for AI factories and support the accelerating global expansion of AI infrastructure — from frontier model training to agentic and physical AI."
SK hynix is also building factory digital twins using NVIDIA Omniverse, OpenUSD scene optimisation pipelines and the open-source cuOpt route-optimisation engine, with the stated goal of enabling fully autonomous fab operations. The companies are additionally exploring agentic AI workflows that connect digital twins to legacy fab software, allowing AI systems to reason over manufacturing data in real time.
Market context and competitive positioning
The partnership arrives at a moment of intensifying competition in the high-bandwidth memory (HBM) market. SK hynix is the dominant supplier of HBM3e, the current generation used in NVIDIA's H100 and H200 GPU clusters, and the agreement extends that alignment into the next architectural cycle. Samsung and Micron are both investing heavily in HBM4 capacity, making supply-chain lock-in increasingly strategic for both GPU vendors and memory manufacturers.
For NVIDIA, securing a named multiyear commitment from SK hynix reduces the risk of memory bottlenecks constraining its Vera Rubin ramp, a concern that became conspicuous during the H100 era when HBM supply repeatedly lagged GPU production. For SK hynix, the partnership provides demand visibility across a broader platform range — from data-centre accelerators to edge and robotics — diversifying its revenue base beyond hyperscaler server memory.
The use of NVIDIA's own simulation and digital-twin tooling within SK hynix's fabs is also notable from a competitive-moat perspective. Embedding CUDA-X and Omniverse into a supplier's manufacturing workflow deepens switching costs on both sides and positions NVIDIA's software stack as infrastructure for the semiconductor supply chain itself, not merely the systems it ships.
Regulatory read-across
Memory supply chains remain a sensitive area of US–China export-control policy. Both NVIDIA and SK hynix operate under US Bureau of Industry and Security restrictions on advanced semiconductor exports to certain destinations; SK hynix, headquartered in South Korea, has navigated these controls as a key supplier of HBM to US hyperscalers. No disclosure was made regarding how the partnership interacts with those restrictions or whether any of the co-developed memory will be subject to additional licensing requirements.
The announcement did not include customer deployment timelines, capacity commitments, or pricing structures. Investors and enterprise buyers will be watching for Vera Rubin system availability dates and for SK hynix to confirm HBM4 yield milestones as the primary near-term signals of whether this partnership translates into deliverable supply.