MIT researchers develop tool to estimate energy use of AI workloads

Researchers from the Massachusetts Institute of Technology and the MIT-IBM Watson AI Lab have developed a rapid estimation system that calculates the energy consumption of AI workloads in seconds, offering a major improvement over traditional methods that take hours or days.

The tool, known as EnergAIzer, is designed to support data centre operators as AI demand accelerates and electricity consumption rises. With AI infrastructure expected to account for a significant share of US power usage in the coming years, more efficient resource planning has become increasingly critical.

EnergAIzer analyses repeatable workload patterns and GPU behaviour to generate fast predictions of energy use across different hardware setups. After incorporating real GPU measurements, the system achieves high accuracy while remaining lightweight and adaptable to current and future chip designs.

By providing immediate feedback on energy consumption, the tool allows developers and operators to optimise workloads, reduce waste, and test different configurations before deployment. The approach is positioned as a practical step towards improving sustainability across large-scale AI systems.

Why does it matter? 

Energy use is becoming one of the defining constraints of AI growth, as large-scale models push data centres towards unprecedented electricity demand. A tool like EnergAIzer directly addresses this bottleneck by making power consumption visible and measurable before deployment.

Faster and more accurate estimation changes how AI systems are designed and scaled. Rather than reacting to energy costs after deployment, developers and operators can optimise workloads in advance, cutting waste and improving efficiency.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!