Sustainable AI is not a destination, it is an operating model. Our previous article explored the challenges of AI’s environmental impact: while AI has the potential to drive efficiencies and reduce environmental damage, it increasingly requires resources, particularly in training large models. But what concrete steps should be taken to integrate sustainability into AI strategies? Here, we move from identifying challenges to providing a practical roadmap for businesses to successfully integrate sustainable practices into their AI operations.
To put it simply: to implement sustainable AI, we recommend two major steps:
- Governance with accountability. Define clear ownership, set environmental budgets, and build checkpoints into the AI lifecycle to ensure impact is measured, reviewed, and addressed.
- Technical integration at each layer. Apply sustainability principles across infrastructure, model design, and data practices through carbon-aware planning, efficient architectures, and leaner data pipelines.
Sustainable AI governance
Governance goes beyond drafting policy documents: it ensures that environmental impact is visible, measurable and integrated into decision-making throughout the AI lifecycle. Effective sustainable AI governance must create clear lines of accountability, set enforceable limits, and integrate sustainability into operational workflows rather than treating it as a separate or secondary concern.
Sustainability should be a fundamental design principle in AI, alongside safety, cost and performance. When integrated into workflows, it becomes a strategic tool for identifying inefficiencies, reducing computational waste, and improving system resilience. Real-time information on emissions and energy consumption enables smarter trade-offs, which often results in simpler models, targeted use of data and better outcomes. Governance adds value by integrating environmental responsibility into daily decisions.