Page 101 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

14 Navigating the intersect of AI, environment and energy for a

sustainable future

In an era in AI is rapidly reshaping industries, understanding its environmental impact and energy Part 2: Thematic AI
dynamics becomes paramount for steering towards a sustainable future. This multidisciplinary
workshop aimed to unravel the complex relationship between AI, the environment, and
energy consumption; spotlight innovations driving AI environmental efficiency; explore AI’s
transformative potential for environmental efficiency in several sectors; and deliberate on the
pivotal role of standards, policies, and regulations.

This event aligned with ITU’s Green Digital Action initiative, reinforcing ITU’s commitment to
promoting digital innovation, standardization, and global collaboration to foster sustainable AI
development while ensuring the ICT sector minimizes its environmental impact and maximizes
its transformative potential.

Presentations for this workshop can be accessed here.

14�1 Understanding AI's environmental impact

This session provided a comprehensive examination of the environmental footprint of AI
systems, emphasizing the resource intensiveness of the AI system lifecycle from data collection
and model training to deployment and inference.

Research from Harvard University on the environmental impact of hyperscale data centres in
the US found that 403 such data centres are responsible for more than 52 million metric tons
of CO₂ annually, representing 1.1 per cent of US electricity-related emissions in 2023. The
carbon intensity of hyperscale data centres is 52% higher than the US average, highlighting the
need for targeted, location-specific strategies. The disproportionate effects on marginalized
communities were also highlighted, underlining the need for data-driven planning tools that
integrate public health and equity concerns. A call was made for a holistic approach that goes
beyond CO emissions to also account for other factors such as air quality.
2
Nvidia highlighted significant energy efficiency gains across its AI stack. For example, LLM
inference energy efficiency has improved 100,000 fold over the past decade, while liquid
cooling systems have reduced water usage by a factor of 300. AI contributes to electricity load
growth but still represents a small share (~0.3%) of global electricity use and is increasingly
powered by clean energy, according to IEA World Energy Outlook 2024 and IEA Energy & AI
Report 2025.

The importance of standards and sectoral cooperation to mitigate AI’s environmental and
material impacts through the standardization work of ITU-T Study Group 5 (Environment,
electromagnetic fields, climate action, and circular economy) were shared. Work addressing
the need for a stable and standardized methodology for measurement include:

• Draft Recommendation ITU-T L.1472 on GHG emissions and energy consumption of the
ICT sector database and pilot project on the implementation of the standard
• Draft ITU-T L.EnvAI guidelines on AI systems environmental impact
• Guidelines for cities to achieve carbon Net Zero through digital transformation

89