As A.I. adoption accelerates, companies are making sustainability claims they cannot reliably substantiate, using metrics that fail to adequately capture the true environmental costs of A.I. systems. This creates a new risk: even well-intentioned companies committed to sustainability may unintentionally greenwash their activities, undermining public trust, investor confidence and corporate credibility.
A.I.’s environmental impacts are significant. They include rising energy demand, increased carbon emissions, water consumption, local environmental degradation and unequal distribution of environmental benefits and harms. Despite growing attention to these issues, the scale of A.I.’s environmental footprint remains difficult to quantify. Much of the A.I. infrastructure is owned by private entities that disclose little about energy use, while technological developments mean existing impact measurement approaches struggle to keep pace. The combination of opacity and rapid change has created a widening gap between what organizations adopting A.I. report about their environmental impact, and what it costs the environment.
The environmental costs of A.I.
A.I.’s environmental impacts span the entire technology stack, from the extraction of raw materials for hardware, through model training and deployment, to the operation of data centers. Data centers have become a flashpoint in the debate about A.I.’s climate impact. They are energy-intensive, high in emissions and strain local grids and infrastructure.
A.I.-specific data centers are projected to consume over 12 percent of U.S. electricity by 2028, with just five OpenAI data centers expected to consume as much electricity as three million homes. However, even figures like these rely on assumptions about usage patterns, efficiency and scale. Without greater transparency from the companies developing and operating A.I. infrastructure, most estimates remain uncertain.
For organizations making sustainability and climate commitments, this uncertainty presents a governance challenge. Claims about sustainability that cannot be verified amidst rising A.I. usage expose companies to reputational risk as scrutiny from regulators, investors and the public intensifies.
The A.I. sustainability tool landscape
A range of tools can help organizations assess the environmental impact of A.I. systems. These include carbon footprint calculators (e.g., Code Carbon), energy consumption and efficiency measurement tools (e.g., NVIDIA Power Capture Analysis Tool (PCAT)), energy optimization software (e.g., Zeus) and life-cycle assessment (LCA) methodologies adapted for A.I. Used well, these tools raise awareness, enable monitoring and enable teams to identify inefficiencies in A.I. workloads. Many are accessible, easy to use and can be integrated into enterprise technology stacks.
However, most of these tools concentrate on narrow windows of the A.I. lifecycle, often focusing on energy use while overlooking the broader impacts of A.I., such as hardware emissions, water usage and localized environmental effects. User feedback highlights other shortcomings, including weak correlation with real-world impacts, gaps in hardware profiling, and difficulties scaling assessments across systems. Few of these tools reflect the full complexity of A.I.’s climate impact or provide the level of detail needed to support in-depth sustainability reporting.
The case for integrated impact assessments
A.I.’s environmental impacts are distributed across geographies and supply chains, and they change over time as models are updated. In the face of this complexity, an integrated approach to assessing A.I.’s environmental impact is required.
Rather than treating sustainability tools as standalone solutions, organizations need to embed them within broader A.I. impact assessment frameworks that consider environmental effects alongside social, ethical and governance factors. Collaborative A.I. impact assessments can help organizations evaluate a wider range of environmental aspects, including pollution, resource consumption, localized effects on communities and impacts on future generations. When aligned with sustainability reporting standards and regulatory requirements, these assessments offer a more credible basis for environmental decision-making and climate commitments.
This approach recognises that a few metrics, like energy efficiency alone, are not a proxy for holistic sustainability—A.I. systems could be optimized for energy use while still contributing to environmental harm in other parts of their lifecycle.
Responsible A.I. or token environmentalism?
Responsible A.I. requires that A.I. is beneficial to humanity, and is safe, ethical and trustworthy. Environmental sustainability is an essential component of this ambition. As A.I.’s environmental impacts grow, organizations cannot afford to ignore existing assessment tools because they are imperfect. At the same time, using these tools without an awareness of their limitations risks unintentional greenwashing and undermines sustainability commitments.
To drive real change and have value, tools and methodologies must be rigorous, accessible, transparent in their assumptions and generate actionable insights. Most importantly, they must be integrated across the A.I. lifecycle rather than applied inconsistently or retrospectively as a reporting exercise.
A.I. will continue to reshape business operations in unprecedented ways. Whether it also undermines corporate sustainability commitments depends on how seriously organizations take the challenge of measuring and mitigating their environmental impacts. Those that actively address this gap will be better positioned to maintain trust and public confidence as scrutiny grows.
Rowena Rodrigues, PhD, is the Head of Innovation and Research Services – Strategic Partnerships at Trilateral Research, a U.K. and Ireland-based SME focused on Responsible A.I. across domains, including climate, energy, health and public safety. Rowena’s expertise spans A.I. governance and impact assessment of new and emerging technologies. Rowena is the co-author of Ethics of Artificial Intelligence: Case Studies and Options for Addressing Ethical Challenges and co-editor of Privacy and Data Protection Seals. She has published widely in leading journals and is well-cited in the areas of A.I., impact assessment and ethics in research and innovation.
Amelia Williams is a Senior Research Impact Officer at Trilateral Research with expertise in scientific communication at the intersection of emerging technologies, environmental issues, ethics and policy. At Trilateral, she supports the development and implementation of research projects alongside policy, media and industry engagement.
