Korean startup develops industrial AI models for data centers and power plants
Trillion Labs, a Korean large language model startup, announced it is building "Industrial World Models" designed to help AI systems understand and optimize complex industrial operations. The company said on June 8 that it will base this work on Nvidia platforms.
These models target environments like AI data centers and power plants. They would analyze engineering drawings, time-series data, equipment records, maintenance histories, and operational constraints to identify optimization opportunities across interconnected infrastructure systems.
Technical approach
Trillion Labs is using Nvidia Omniverse for physics-based simulation and Nvidia Nemotron models for reasoning capabilities. The company previously developed foundation models including large language models and vision-language models from scratch, then shifted focus toward world models that can simulate physical environments.
The startup is developing AI models optimized for real industrial settings, covering the full lifecycle of model training, inference, deployment, and operations.
Market positioning
Trillion Labs frames this work as the foundation for a new "Industrial Intelligence" sector. The company plans to combine world models with AI factories, digital twins, and physical AI ecosystems.
CEO Shin Jae-min said the next phase of AI requires "understanding and reasoning about the real world," not just language processing. The company is pursuing partnerships with various organizations including affiliates of GS Group to explore energy and industrial infrastructure applications.
For IT and development professionals, this represents a shift toward AI systems that can model and optimize the infrastructure these systems run on-a practical extension beyond current language and vision capabilities.
Your membership also unlocks:
