NVIDIA and SK hynix announce a multiyear deal to co-develop next-gen memory for AI factories, supercomputers, PCs, and robots.
For years, the relationship between NVIDIA and SK hynix has operated quietly in the background of some of the most powerful AI hardware ever built. On Saturday, the two companies made that relationship official and significantly more ambitious, announcing a multiyear technology partnership designed to keep memory supply in step with the accelerating global buildout of AI factories.
The announcement, made jointly by both companies, covers far more than procurement. It is a deep technical collaboration that spans memory architecture, semiconductor simulation, and factory automation — a signal that the infrastructure race powering the current AI cycle now extends to every layer of the supply chain.
"AI factories are the engines of the next industrial revolution, and advanced memory is essential to their performance."Jensen Huang — Founder and CEO, NVIDIA
Jensen Huang has spent years describing the shift in how computing resources are being deployed — away from general-purpose servers and toward specialized, continuously operating facilities that train frontier AI models and run inference at scale. In that framing, memory is not a component; it is a critical throughput bottleneck. More bandwidth and greater capacity directly translate into faster model training and more responsive AI applications.
SK hynix is one of the world's two largest producers of High Bandwidth Memory, the type of memory that sits directly beside the processors inside NVIDIA's flagship data-center GPUs. The partnership formalizes a co-engineering arrangement that, according to both companies, has been building for years.
"SK hynix and NVIDIA have been building toward this for years, and this partnership reflects the depth of that collaboration. Together, we are co-developing the next generation of memory for AI factories and applying AI to how we design and manufacture semiconductors — work that will shape the future of AI infrastructure."
Chey Tae-won — Chairman, SK GroupExpanding Into New Markets NVIDIA Is Creating
One of the more striking elements of the deal is its breadth. Rather than focusing solely on data-center hardware, the agreement extends SK hynix's footprint into every major product category NVIDIA is building toward: large-scale AI infrastructure, personal AI, and physical AI.
The reach of the agreement reflects the degree to which NVIDIA has transformed from a graphics chip company into an end-to-end AI computing platform. Each of those product lines carries substantially different memory requirements — different bandwidth profiles, power constraints, and physical form factors — making the co-engineering commitment significant in both scope and technical complexity.
AI Infrastructure
Memory for NVIDIA Vera Rubin AI supercomputers and Vera CPUs — the backbone of large-scale model training and inference.
Personal AI
Memory optimized for NVIDIA RTX Spark-powered PCs — the consumer devices at the center of NVIDIA's local AI ambitions.
Physical AI
Memory designed for NVIDIA Jetson Thor robotic computing platforms — the hardware underpinning the next generation of autonomous machines.
AI Enters the Semiconductor Design Lab
Beyond the hardware partnership, one of the most forward-looking aspects of the deal involves applying AI to the process of making chips in the first place. SK hynix is using NVIDIA's CUDA-X libraries and the PhysicsNeMo framework to accelerate technology computer-aided design — the dense, physics-intensive simulation work that underpins how new semiconductor devices are modeled before fabrication.
Computational lithography, another area where simulation cycles are notoriously long and computationally expensive, is also part of the scope. By extending these tools into SK hynix's in-house engineering workflows, the companies say they are opening a pathway for three-way collaboration among chipmakers, NVIDIA, and electronic design automation software vendors — a corner of the semiconductor industry that has historically operated in silos.
The practical effect, if it materializes as described, would be a measurable compression of the development cycle for new memory architectures. Faster simulation means faster iteration. And faster iteration is exactly what the AI hardware cycle currently demands, with model architectures and compute requirements evolving at a pace that strains traditional memory roadmaps.
Building a Digital Mirror of the Semiconductor Factory
Perhaps the most ambitious thread in the partnership involves what happens inside the fabrication facility itself. SK hynix is developing factory digital twins — detailed, real-time virtual replicas of its manufacturing environments — using NVIDIA Omniverse and OpenUSD pipelines as the underlying platform.
These are not visualization tools in any conventional sense. The goal is to simulate and optimize the movement of autonomous mobile robots, track fab assets, and model complex manufacturing processes within a three-dimensional environment that can be updated continuously as the physical facility changes.
To support logistics optimization within those environments, SK hynix is using NVIDIA cuOpt, an open-source GPU-accelerated decision engine that handles routing and scheduling problems of the kind that arise constantly in a high-throughput semiconductor plant. The NVIDIA Metropolis platform, which handles intelligent video analytics, is also part of the stack.
The two companies are also in early discussions about connecting these digital twins to agentic AI systems — workflows in which AI agents reason over live factory data, automate routine tasks, and improve manufacturing decisions without waiting for human input. That kind of autonomous operation is still a future state rather than a present reality, but the infrastructure being built today is explicitly designed to support it.
Taken together, the three pillars of the partnership — memory co-development, semiconductor simulation, and factory digital twins — sketch a picture of how deeply the two companies plan to integrate their engineering and product roadmaps. It is a bet that AI will not only drive demand for new hardware but will also become the primary tool for designing and building that hardware faster and more efficiently.
The timing of the announcement is notable. It comes as AI infrastructure investment continues at an extraordinary pace globally, with data center buildout generating sustained demand for advanced memory well beyond what a transactional supplier relationship could reliably serve. Multiyear agreements of this depth are, in part, a hedge against supply uncertainty — a way for both sides to plan capital investment and engineering resources against a shared roadmap rather than annual purchase orders.
For SK hynix, the deal is also a chance to diversify. While the company has long been a critical supplier to the data center market, the extension into personal AI PCs and robotic platforms represents meaningful new revenue streams — categories where NVIDIA's platform is just beginning to define what "adequate" memory looks like.
NVIDIA's stock trades on the NASDAQ under the ticker NVDA. SK hynix shares are listed on the Korea Exchange, with Global Depository shares on the Luxembourg Stock Exchange.
About SK hynix: Headquartered in South Korea, SK hynix is one of the world's largest semiconductor companies, specializing in DRAM and NAND flash memory. Its products are used across a broad range of consumer electronics, servers, and AI hardware platforms worldwide.
About NVIDIA: NVIDIA (NASDAQ: NVDA) is the world leader in AI computing and accelerated processing. The company designs graphics processing units, system-on-chip units, and AI software that power data centers, autonomous vehicles, robotics, and consumer devices.
0 Comments