Nvidia Blackwell AI Factory South Korea 2025

Avatar
Lisa Ernst · 31.10.2025 · Technology · 5 min

South Korea is developing into a global center for AI infrastructure. A consortium of government sovereign cloud and enterprise-driven AI factories with over 260,000 Nvidia GPUs is forming. Projects at Hyundai Motor Group, SK Group and Samsung are driving this development forward.

Introduction

An AI Factory is a data center that converts data into intelligence. Training, simulation, validation, and operation of AI models occur as a closed production process, comparable to a factory. Blackwell is Nvidia's current GPU architecture for data centers and high-performance AI. It drives the new factories, including DGX systems for training, RTX PRO servers for simulation/digital twins, and platforms for robotics. Souveräne KI-Infrastruktur refers to national or regional computing capacity in local clouds and partner data centers to reduce dependencies and safeguard privacy/compliance.

Current status

South Korea is building a national AI infrastructure with more than 260,000 GPUs in collaboration with Nvidia. The government plans over 50,000 of the latest GPUs for a government AI computing consortium. Companies like Samsung, SK Group and Hyundai are each constructing factories with around 50,000 GPUs. NAVER expands to over 60,000 GPUs. Hyundai Motor Group und Nvidia are establishing a Blackwell-based AI Factory for autonomous mobility, Smart Factories and robotics. The goal is integrated training, validation and deployment with 50,000 Blackwell GPUs and roughly a 3-billion USD boost for the ecosystem. SK Group is building an AI Factory with more than 50,000 GPUs, with its first expansion phase due by the end of 2027. It supports digital twins and agent development. Samsung plans with Nvidia an AI Factory with over 50,000 GPUs for 'agentive' and 'physical' AI in semiconductor manufacturing, mobile devices and robotics, including Omniverse-based digital twins for global fabs. Reuters frames the agreements politically: announcement on the sidelines of the APEC meeting in Gyeongju, involvement of the largest Korean conglomerates and a focus on manufacturing, mobility and cloud capacities.

A futuristic depiction of a NVIDIA AI Factory, showcasing the capabilities of the Blackwell architecture for AI applications.

Source: blogs.nvidia.com

A futuristic depiction of a NVIDIA AI Factory, demonstrating the capabilities of the Blackwell architecture for AI applications.

Analysis

Three forces are at work: sovereignty over AI capabilities, scaling 'agentive' AI in real-world processes, and industrial digitization. APEC provides the stage for multilateral commitments. Für die Industrie They are driven by shortened development cycles through digital twins, validated autonomy in simulated extreme scenarios, and on-device intelligence in vehicles and robots. Für Nvidia Korea is an attractive market beyond Chinese export restrictions. The deals spread risk and unlock new application fields in manufacturing and trade. Hyundai considers the AI Factory as a 'control center' across the entire lifecycle from model training to fleet operation. Samsung connects the AI Factory with semiconductor OPC, EDA workflows and global factory twins. SK also builds an industrial cloud on RTX PRO 6000 Blackwell servers for startups, businesses, and authorities.

Source: YouTube

The GTC-Keynote-Mitschnitt shows Blackwell, AI Factories and the bridge to robotics and digital twins firsthand.

Facts & Reactions

Over 260,000 GPUs are confirmed for Korea, including >50,000 for the government, about 50,000 each for Samsung, SK Group and Hyundai, and >60,000 for NAVER. The APEC framework and the goals in mobility, manufacturing and telecommunications are also documented. Die Hyundai-AI-Factory with 50,000 Blackwell GPUs and around 3 billion USD of stimulus focused on autonomous driving, Smart Factory and robotics. Die SK-AI-Factory with >50,000 GPUs is planned in the first phase by the end of 2027. Unklar are exact locations, energy concepts and network connectivity of the individual factories. The press materials do not specify detailed data center sites or PUE targets. Liefer- und Ramp-Up-Zeitpläne per project details are also unclear. Reuters cites the total scope, but no precise quarterly milestones. Die Annahme , AI-Factories are only 'large training computers', which is misleading. The documents describe the end-to-end network from training, simulation, validation and operation including digital twins and robotics. Nvidia-CEO Jensen Huang frames AI Factories as new 'production facilities' of the AI economy. SK-Chairman Chey Tae-Won emphasizes the role for storage, Digital Twins and agents. Regierungsseitig is positioned as a growth engine and sovereignty project for the national AI infrastructure. Reuters also points to geopolitical frameworks and the diversification of markets.

Infrastructure preparation: Server racks are delivered for installation in an NVIDIA AI Factory.

Source: blogs.nvidia.com

Infrastructure preparation: Server racks are delivered for installation in an NVIDIA AI Factory.

Impact & Conclusion

For manufacturers, mobility providers and robotics teams, a fast track to validated autonomy opens: data from fleets and factories land in the AI Factory, generating new software versions in simulation and digital twins, which then flow back into production and vehicles. Zulieferer should assess how Omniverse-based twins and GPU-EDA chains can be integrated into existing tool landscapes. Für Startups in Korea On-ramps arise via industrial cloud offerings from SKT, including GPU-as-a-Service and access to ecosystems such as Nvidia Inception. Tipp : Check the primary source (Newsroom/IR) for every large figure and compare it with independent media reports, especially for investment amounts and timelines. Offene Fragen affect energy and grid expansion needs as well as the anchoring of sustainability targets (PUE, CO2 factor). The press releases do not specify measurable environmental KPIs. Die Geschwindigkeit , with security and validation processes for autonomous functions at fleet level certified, is also unclear, as regulatory roadmaps are missing from the material. Abhängigkeiten von Lieferketten (e.g., HBM) and their safeguarding are further open points. There are references to memory cooperation, but no quantified risk buffers. Nvidia und südkoreanische Konzerne drive industrial AI forward: AI Factories become an infrastructure class that connects mobility, factories and robotics – with documented projects in six-figure GPU scale and clear roles for state and corporations. Wer Produkte oder Prozesse operating with data can close the gap between simulation and reality—if capacity, data quality and validation are planned early and primary sources are kept in view.

Source: YouTube

The NVIDIA Blackwell GPU board, the centerpiece of the new generation of AI data centers and AI Factories.

Source: whileint.com

The NVIDIA Blackwell GPU board, the centerpiece of the new generation of AI data centers and AI Factories.

Share our post!