NVIDIA Can't Take It Anymore! NVIDIA Partners with Samsung to Develop In-House NAND, Will Memory Manufacturers Focused on Capacity Be Eliminated?

Many people are unaware that the role of the memory industry is undergoing a fundamental transformation. In the past, GPU giant Nvidia was just a major memory purchaser, but now it is gradually moving into a deeper position in the supply chain: actively participating in the design of next-generation storage chips. According to Korean media reports, Samsung has established a partnership with Nvidia to jointly develop the next generation of NAND Flash memory. Future memory will no longer be standardized commodities but core components tailored for AI architectures.

(What changes has Nvidia Vera Rubin brought? Analyzing the Memory War Era: SK Hynix, Samsung, Micron, SanDisk)

Nvidia Accelerates Chip Development Speed by 10,000 Times Using AI Models

Sources indicate that Samsung Semiconductor Research Institute, Nvidia, and Georgia Institute of Technology research teams have jointly developed an AI model called Physics-Informed Neural Operator (PINO). This system can analyze the performance of new NAND memory devices, achieving speeds over 10,000 times faster than traditional simulation methods.

In semiconductor R&D, engineers typically rely on a simulation tool called TCAD (Technology Computer-Aided Design) to test chip designs. However, this method usually takes about 60 hours per run, severely limiting R&D efficiency. The research team combined physical laws with neural networks, enabling AI to understand the physical behavior of materials and devices, reducing simulation time to less than 10 seconds. This achievement has been published in the international research community.

Samsung’s Secret Weapon: Ferroelectric NAND

The core technology of this collaboration is a new storage technology called ferroelectric materials. Ferroelectric materials can maintain their polarized state of positive and negative charges even without continuous power input, making them suitable for data storage with extremely low power consumption. Samsung has been a leading researcher in this field. By the end of 2025, Samsung published a study in the journal Nature indicating that ferroelectric NAND could reduce power consumption by about 96% compared to traditional NAND.

This means that future AI systems accessing large-scale data will significantly reduce energy consumption.

Memory Manufacturers That Only Expand Capacity Without Improving Technology Will Be Eliminated?

This collaboration signifies that the AI memory race has entered a new stage. In the past, Nvidia mainly supported AI GPUs by purchasing high-bandwidth memory (HBM) and NAND. Now, it has decided to develop dedicated memory chips. As AI model sizes grow explosively, storage architecture itself is becoming part of AI system design. Perhaps future memory chips will no longer be just general-purpose components but “architectural plugins” tailored for AI systems.

According to the Korean Intellectual Property Office (KIPO), currently, Korea holds 43.1% of global ferroelectric-related patents, with Samsung alone accounting for 27.8%.

This article, “Nvidia Can’t Take It Anymore! NVIDIA Partners with Samsung to Develop Self-Designed NAND, Memory Manufacturers Focused Only on Capacity Expansion Will Be Eliminated,” first appeared on Chain News ABMedia.