The Energy Challenge of AI
DNV’s 2025 Energy Transition Outlook highlights AI as a significant new source of global electricity demand
Doubling by 2030
Data-centre power use will more than double by 2030, driven by training and deploying large-scale AI models
Hardware vs. Workload
Although hardware efficiency improves by ~40% per year, total energy consumption continues to rise due to exponential workload growth
Architecture Matters
The energy footprint depends on model size, data efficiency, and architecture choices — not just hardware improvements
3LC as an Energy-Transition Enabler
Decoupling AI performance from energy growth through smarter data and efficient architectures
Smaller, more efficient models
3LC consistently delivers 30–40% accuracy improvements on existing datasets while using up to 90% less data. This reduces GPU training time by 2–3× and electricity demand during training. In production, smaller models mean lower energy per inference and longer hardware life.
Data reduction and model efficiency proven in global competition
In the Nexar Autonomous Driving Challenge, 3LC CEO Paul Endresen won Kaggle’s largest dashcam competition using only 3LC. The winning model achieved 0.898 (vs. 0.71 baseline) using just 2–5% of the data provided and without any parameter tuning. Because 3LC identified only the most informative samples, the final model was 1/30 the size of the runner-up’s — yet achieved higher accuracy. This proves that smarter data, not bigger models, drives real-world performance while dramatically cutting energy use.
Reduced data movement and storage
3LC operates entirely on-premise, eliminating large-scale cloud transfers and associated network energy cost. Clients like Equinor report a 1/3 reduction in data-science workload, freeing up teams without expanding compute infrastructure.
Fewer retraining cycles and less waste
In the eSmart Systems case, 3LC achieved 40× faster labeling and cleaner data, cutting retraining cycles that typically consume thousands of GPU-hours per year.
Edge-ready, domain-specific AI
Compact models produced through 3LC can be deployed on robots, drones, or inspection units — aligning with the world’s need for distributed, energy-efficient AI architectures and reducing dependence on hyperscale data centres.
AI Meets the Energy Transition
AI’s rising electricity footprint is not only a technology issue — it is a central part of the global energy transition. 3LC helps organizations decouple AI performance from energy growth, achieving better results with less data, smaller models, and lower compute intensity. Each 3LC deployment reduces GPU hours, cooling load, and data movement — making AI cleaner, cheaper, and faster.
In short, 3LC is not just a productivity tool — it is an energy-transition technology, enabling industries to accelerate AI adoption while staying within planetary and grid limits.