[Exclusive] Could Samsung and SK hynix Lose Their Know-How With a Single U.S. Law? The 'Sleeping With Nvidia' Dilemma Facing K-Semiconductors

[The Two Faces of Digital Twins] ①
Samsung and SK hynix Accelerate Toward ‘Autonomous Factories’ with Nvidia Technology
Concerns Over Process Data and Know-How Exposure Despite ‘Closed Networks and NDAs’
“Without Technological Independence,

Korean semiconductor companies have found themselves in a dilemma. While adopting the Nvidia platform is inevitable to enhance manufacturing competitiveness, doing so could expose data containing decades of accumulated process know-how to external platforms. Samsung Electronics and SK hynix are implementing Nvidia’s core platforms in their semiconductor manufacturing processes and have put in place dual safeguards against technology leakage, such as establishing closed networks and signing non-disclosure agreements (NDAs). However, experts warn that individual company defenses alone are insufficient to protect process data, which is directly linked to national security. They point out that, due to structural loopholes that allow data to be accessed at any time under U.S. law, it is urgent for the nation to establish infrastructure and policy strategies to secure sovereignty over industrial data.

The Two Giants of Memory Accelerate Their Shift to AI Factories

As of July 6, according to the semiconductor industry, Samsung Electronics and SK hynix are transitioning to running the entire semiconductor production process on Nvidia technology, from process design to operation and improvement. This goes beyond simple automation, as they are reorganizing their manufacturing paradigm by applying artificial intelligence (AI) and digital twin technology at all stages, including process design, operation, and optimization.

The core tool is the digital twin. This technology replicates an actual factory in a three-dimensional virtual space, allowing processes to be simulated in advance to find optimal conditions. Samsung Electronics has implemented digital twins of major production facilities, such as Pyeongtaek Plant 1, using Nvidia's virtual simulation platform, Omniverse. Through this, the company is advancing an integrated control system that analyzes equipment status in real time and detects anomalies. It is reported that Samsung has also adopted Nvidia’s computing program, cuLitho, for the photolithography process that etches semiconductor circuits onto wafers, improving design accuracy and reducing development time. The plan is to expand the use of digital twins to overseas production sites, including the Taylor plant in the United States.

SK hynix has also begun building a factory digital twin using Omniverse, completing related proof of concept (PoC) testing last year. By utilizing Nvidia’s development tools CUDA-X and PhysicsNeMo, the company is also increasing the speed of design and manufacturing simulations. The goal is to establish an autonomous fab (factory) that learns and makes decisions on its own without human intervention by 2030.

The Paradox of Cost-Effectiveness: A Structure of 'Voluntary Dependence'

The problem lies in the difficulty of exiting once a company enters this ecosystem. Nvidia offers everything needed for a smart factory—from graphics processing units (GPUs) to simulation platforms and AI development tools—as a single package. Once a company adopts this package, all data formats and work processes in the factory become tailored to Nvidia’s standards. In the future, even when introducing new equipment or changing processes, handling everything within the established Nvidia environment becomes the cheapest and fastest option. Switching to another company’s platform would entail enormous costs and the risk of halting factory operations during the transition.

Nvidia has succeeded with this approach before. In the AI development field, the structure in which using GPUs inevitably leads to using Nvidia’s development tool CUDA has solidified over several years, making CUDA the de facto global standard for AI development. If the same strategy succeeds in the manufacturing sector, there are concerns that companies, in pursuit of performance, could become voluntarily locked into Nvidia’s ecosystem.

Kim Myungjoo, Professor of Information Security at Seoul Women’s University and Director of the Artificial Intelligence Safety Research Institute, pointed out, “Nvidia is employing the same strategy in the digital twin and physical AI domains as it did when monopolizing GPUs in the past. Although Korean companies are aware of the risks, their options are limited because there is currently no suitable alternative technology or platform that can match the sophistication of these systems.”

Closed Networks and NDAs in Place... "Still Cannot Guarantee Safety"

Of course, companies are not standing by idly. Samsung Electronics operates Omniverse-based digital twin services within a physically separated, closed private network and signs NDAs with platform providers at the contract stage. SK hynix completely separates key process data from external frameworks for independent management, and its core solutions for production, logistics, and physical simulations are run on its own platform, with Omniverse used solely as a subordinate tool for digital twin implementation. The intent is to fundamentally block the possibility of core technology leaks, even while adopting external platforms.

However, experts widely agree that such measures alone cannot guarantee safety. When digital twins and physical AI are combined, the structure shifts to one where AI learns in real time from on-site data, and in this process, on-site data—including process know-how—can be learned by the external platform in reverse. In other words, the manufacturing competitiveness of Korean semiconductors, accumulated over decades, could be siphoned off via the platform.

Moon Songcheon, Professor Emeritus at the KAIST Graduate School of Management, commented, “There is a tendency to believe that inserting data leakage prevention clauses in contracts and building on-premises (closed) servers on-site makes things safe, but one must not overlook that if the software provider wishes, they can access the data at any time. The core technologies for software engines, operating systems (OS), database management systems (DBMS), and network communication protocols running inside closed servers are still controlled by big tech companies.”

There are also legal loopholes. Choi Byungho, Professor at the Korea University AI Research Institute, pointed out, “Regardless of the terms and NDAs, if the location of the cloud for platform integration is in the United States, U.S. authorities can access the data.” According to the U.S. Cloud Act, in the event of a national security crisis or at the request of the U.S. government, data stored on U.S. cloud servers can be legally accessed by force. Professor Choi suggested, “Just as Google once announced it would build a data center in Korea to protect data sovereignty, there must be sophisticated negotiations with Nvidia to keep data centers in Korea and encrypt the data.”

Professor Kim also advised, “When building digital twins, if sensor or camera information from the real world is linked indiscriminately, unnecessary company know-how may be exposed externally. It is necessary to use technology that minimizes training data so that only essential information is filtered for AI learning, and to stipulate in the contract at the negotiation stage that certain data will be excluded from training.”

"National Governance Beyond Corporate Defense Is Urgently Needed"

In the long term, the establishment of national-level governance is emerging as a challenge. Only a few companies, like Samsung Electronics and SK hynix, have the capacity to build their own security infrastructure. Most medium-sized and small manufacturing firms are being incorporated into the global platform ecosystem without separate data protection systems.

Professor Kim emphasized, “For most companies, immediate profit takes precedence, making it difficult to see the risks of dependence that may arise 10 to 30 years down the line. The government must play a governance role—monitoring the industrial ecosystem in the mid-to-long term and establishing research and development (R&D) alternatives.” Professor Moon also warned, “During a semiconductor supercycle, the seriousness of dependence is masked, but if we do not achieve self-reliance in core software technologies in the long term, the foundation of the industry itself may be shaken after the supercycle.”

A high-ranking semiconductor industry official also stated, “Design, process, and production data are core assets of the semiconductor industry. It is urgent for the government to expand domestic AI computing infrastructure and data centers, establish a secure data sharing system, and provide institutional and technological support for security technologies that meet international standards.”

Possible strategies include domestic development of world models and the establishment of a state-led physical AI ecosystem. A world model is an AI that learns the principles and rules by which the real world operates as a whole. With such a model, various environments—such as actual factories or roads—can be virtually recreated to simulate and plan any operation in advance. Professor Choi explained, “Advanced world models require massive data and sophisticated manufacturing technology. The United States could not achieve unilateral dominance in this field due to a lack of local manufacturing infrastructure and productivity. Korea’s world-class manufacturing technology and process data can become powerful weapons.”

This content was produced with the assistance of AI translation services.