The relationship between artificial intelligence and energy is not linear but rather a reinforcing loop. AI is reshaping how energy systems are planned, operated, and optimized. At the same time, the rapid growth of AI is making energy a strategic constraint rather than a passive input. Understanding both directions, AI-for-Energy and Energy-for-AI, and how they interact is essential for policy design.
In the AI-for-Energy direction, AI has already delivered measurable value across the energy sector by improving demand forecasting, optimizing asset management, and supporting smarter engineering operations. The next frontier is generative AI (GenAI), which moves beyond pattern recognition. Rather than simply mapping inputs to outputs, GenAI learns the underlying distribution of data and can synthesize realistic scenarios. These capabilities are highly valuable for analyzing rare, high-consequence events that are historically underrepresented in operational data. Where earlier tools optimized what was measured, GenAI can synthesize data for what is missing. In KAPSARC's analysis of predictive maintenance, filling this gap with synthetic failure cases raised projected maintenance cost savings by 31%.
Deployment barriers, however, remain significant. KAPSARC's analysis identifies trust, rather than mere accuracy, as the principal constraint on GenAI adoption. The key questions are: Who is accountable for a model's advice? Is oversight adequate? And what happens when models fail? The appropriate policy response is to adopt an adaptive approach: minimal supervision for low-risk applications with no impact on critical operations, certification and monitoring for real-time decisions, and no autonomous control without human override. This graduated approach, calibrated to the risk of each use case, offers a workable template for AI-in-Energy policy.
In the Energy-for-AI direction, the scale of AI's energy appetite is now a planning reality. Our analysis indicates that global data center capacity (for all data center types, excluding crypto mining), about 112 GW IT in 2024, is projected to double to 224 GW by 2030, and the corresponding electricity use is projected to rise from 854 TWh to nearly 1,900 TWh over the same period. Cost must therefore be carefully modeled and managed. KAPSARC's levelized cost analysis finds that the price of AI compute, and by extension the cost per token, is governed more by AI chip generations and utilization rates than by electricity tariffs or cooling efficiency. Each new generation of AI hardware significantly lowers the levelized cost of compute. For example, the cost was lowered by nearly 60% from 2022 to 2025. On the other hand, our analysis indicates that doubling the tariff only raises the cost by around a fifth. The strongest levers are therefore operator decisions. The levers governments hold, including tariff design and efficiency standards, matter less, yet remain relevant for policy.
Data centers also offer a distinctive source of demand flexibility. While some AI workloads remain geographically anchored due to certain requirements such as data sovereignty constraints, other AI workloads, such as insensitive training of large language models, are both computationally intensive and geographically flexible, so they can shift across geographies to absorb variable renewable generation. Siting, interconnection, and market design therefore become central policy instruments.
The loop completes itself. Cheaper, more efficient AI enables deeper grid optimization, accelerates the discovery of clean energy materials, and improves renewable forecasting. Yet this efficiency may also drive higher aggregate energy demand, potentially triggering a classic rebound effect. The question is no longer whether AI and energy are interdependent. The policy challenge is to design frameworks that capture value in both directions simultaneously.
References:
- Generative AI for the Power Sector: Methodology, Safety, Trustworthiness, and Policy Implications
- AI and Energy: The Future of Data Centers in Saudi Arabia
Guest Author
