What does 'sustainable AI' actually mean in practice?

A Broad Concept Without One Fixed Definition

“Sustainable AI” is a term used to describe efforts to reduce the environmental impact of developing and operating AI systems, but it’s important to understand upfront that there isn’t one single, universally agreed technical definition or certification standard governing what qualifies. Instead, it’s better understood as an umbrella term covering several related practical goals, each aimed at making AI development and deployment less environmentally costly than it would otherwise be.

This means that when a company or organization describes its AI efforts as “sustainable,” it’s worth looking at the specific practices and metrics behind that claim, since the term can be applied with varying degrees of rigor and specificity depending on who’s using it.

The Practical Components Behind the Term

In practice, sustainable AI efforts generally fall into a few recognizable categories. Energy efficiency is a major focus, including both making AI models themselves more computationally efficient, so they require less processing power to train and run, and making the underlying hardware and data center infrastructure more efficient in how it converts electricity into useful computing work.

Cleaner energy sourcing is another significant component, referring to efforts to power AI data centers using renewable or lower-emission energy sources rather than relying entirely on more carbon-intensive power generation, where available and feasible given local energy grid conditions.

Water use reduction addresses the water consumption associated with certain data center cooling methods, particularly evaporative cooling systems, with sustainable AI efforts in this area often focused on adopting cooling approaches that reduce or eliminate this water dependency.

Finally, transparency and measurement efforts focus on more accurately tracking and reporting the actual environmental costs of AI development, including energy consumption and emissions associated with training and running AI models, which is itself an ongoing challenge given the complexity of measuring these costs consistently across different organizations and AI systems.

Why the Lack of a Single Standard Matters

Because there isn’t yet one authoritative, universally adopted definition or measurement framework for sustainable AI, different organizations may emphasize different aspects of this broad concept, or measure their progress using different methodologies, which can make direct comparisons between different companies’ sustainability claims difficult. Various industry groups, research institutions, and government bodies have worked on developing more standardized metrics and frameworks, but this remains an evolving area rather than a settled, mature standard.

Bottom Line

Sustainable AI, in practice, refers to a collection of related efforts aimed at reducing AI’s environmental footprint, including improving energy efficiency in both models and hardware, using cleaner energy sources to power data centers, reducing water consumption in cooling, and improving transparency about AI’s actual environmental costs. Because there’s no single standardized definition or certification, evaluating any specific sustainability claim requires looking at the concrete practices and metrics behind it.

Go deeper

• How Do AI Companies Justify Their Environmental Footprint?
• Are There Ethical Frameworks Specifically for Evaluating AI’s Environmental Impact?
• Can the Environmental Benefits of AI Outweigh Its Energy Costs?