South Korea's two industrial giants—LG Group and Doosan Group—are simultaneously announcing major collaborations with NVIDIA to build proprietary AI factories, a development that signals a fundamental shift in how large multinational manufacturers approach compute infrastructure. Rather than relying exclusively on cloud-based GPU resources, both conglomerates are constructing vertically integrated systems designed to train, simulate, validate, and deploy physical AI models across their entire business ecosystems. LG's AI factory will span robotics, autonomous driving capabilities, data center technologies, and GPU cloud services, while Doosan's initiative encompasses Doosan Robotics, Doosan Bobcat (heavy equipment), Doosan Enerbility (renewable energy), and Doosan Corporation Electro-Materials. The move reflects growing recognition among enterprise customers that controlling compute infrastructure in-house reduces latency, improves security, and enables rapid iteration on proprietary models—particularly critical for physical AI and robotics applications where inference speed directly impacts competitive advantage.
Both partnerships leverage NVIDIA's full stack of accelerated computing technologies, though neither announcement specifies which GPU architectures (Hopper, Blackwell, or custom configurations) are being deployed at scale. This omission is notable: it suggests these AI factories may involve custom silicon arrangements or multi-generational hardware strategies that manufacturers prefer to keep confidential. The timing is significant because it coincides with broader enterprise pushback against concentrated cloud computing costs and latency constraints. By building captive AI infrastructure, LG and Doosan gain the ability to train models on proprietary manufacturing, robotics, and supply-chain datasets without sending sensitive data through third-party cloud providers. For NVIDIA, these deals represent high-value, long-term commitments that lock in GPU consumption across the entire product lifecycle of industrial robots, autonomous vehicles, and edge compute systems—markets where inference volumes will dwarf traditional data center workloads.
The South Korean factories represent a microcosm of a larger pattern emerging across global manufacturing and infrastructure sectors. The UK's sovereign AI initiative, announced with Jensen Huang at London Tech Week, similarly aims to build national compute capacity rather than depend on US-based cloud providers. Apple's integration of NVIDIA GPUs into its Private Cloud Compute infrastructure—now expanding to Google Cloud—demonstrates how even the world's largest technology companies are architecting hybrid approaches that balance proprietary compute needs with cloud flexibility. LG and Doosan's announcements suggest that large industrial conglomerates with diverse business units are reaching the same conclusion: owning strategic GPU infrastructure is no longer optional for competitive AI deployment. Neither company has disclosed capex figures or timelines, but these AI factories likely represent multi-billion-dollar commitments that will reshape how NVIDIA's enterprise sales are structured over the next five years, shifting from per-instance cloud consumption toward integrated infrastructure contracts.