How can AI Address the Energy Efficiency Objectives in the Industry?

The rise of generative AI has reignited the debate about the energy cost of AI. However, so far, the potential for AI to help optimize resource use, or to facilitate firms’ transition from dirty to green production technologies, has been largely overlooked. No serious attempt has been made at computing the overall energy cost and/or benefit of AI. Traditional AI models require significantly less energy-intensive training and inference and could enable companies to improve the energy efficiency of their technologies and help them meet their environmental objectives for the coming years. Machine and deep learning models used in industry and deployed using digital twins can control complex processes and optimize their resource use, in energy-intensive industries, and achieve significant energy-saving potential. Wastewater treatment is emblematic in this respect: aeration during secondary treatment typically accounts for roughly half of a plant’s electricity consumption. Our focus in this article is on the comparison between the direct energy cost of AI operations and the energy saving AI induces when implemented by Veolia.

Veolia, in partnership with PureControl, has rolled out one of the first large-scale deployments of AI for climate-relevant efficiency, covering about 200 plants. PureControl’s system ingests high-frequency data (≈15 minutes intervals) on electricity prices, weather, sensor streams, and laboratory quality samples to maintain a live digital replica of the plant. The AI then schedules and doses aeration to minimize cost and consumption while assuring effluent quality and regulatory thresholds.

We evaluate AI’s net effect using plant-level operational data, natural experiments (unplanned interruptions), and a full accounting of the AI layer’s own electricity use. Preliminary results from ~15 plants indicate a nearly 10% reduction in electricity consumption and GHG emissions, while AI’s direct electricity use accounts for less than 1% of the gross energy savings. Even under conservative assumptions about additional required hardware installations, the maximum lifecycle carbon cost of AI remains well below the emissions abatement, thereby pointing to a robust net-positive climate contribution.