Managing AI’s Environmental Impacts

Introduction

As companies move from AI pilots to enterprise-wide deployment, consumers, stakeholders, and regulators are expecting them to understand, measure, and manage AI-related environmental impacts. In response, BSR members are seeking practical guidance on issues ranging from carbon emissions, water use, and land use to supplier engagement, customer requests, governance, standards, and collaborative action.

This FAQ provides practical starting points for sustainability professionals. It is not intended as legal, assurance, or formal GHG accounting advice. AI measurement remains an evolving discipline that will require further standardization and rigor.

How can companies measure the carbon footprint of AI use and AI products? 

We recommend that companies separate two measurement questions: emissions from enterprise AI (AI used internally) and emissions from AI-enabled products or services (AI incorporated into customer-facing offerings).  

• For enterprise use, sustainability teams can work with IT, procurement, product, and cloud teams to identify which AI tools are being used, where they are hosted, what data or models they rely on, and what activity data is available from providers.  
• For AI-enabled products, lifecycle and product-carbon-footprint framing is more appropriate. We recommend that companies consider business model dependencies, cloud and data center operations, hardware, use-phase energy, and downstream customer use. Depending on the company’s role and inventory boundary, these impacts may fall across Scope 2 and multiple Scope 3 categories, including use of sold products. Simple metrics, such as tokens per watt-hour, can be useful for screening, but they should not be substituted for appropriate measurement methods, transparent and credible boundaries, and better supplier-specific data over time. Measurement and transparency remain underdeveloped across the AI value chain, so companies should treat this as a repeatable process that can improve annually with accompanying success metrics. 

How can companies measure broader environmental impacts of AI, including water and land use? 

Companies can begin by looking across the full AI value chain, not only visible use of AI tools. Water impacts may arise from material extraction, semiconductor manufacturing, electricity generation, and data center cooling. Such horizon scanning may also identify land impacts that arise from mining, manufacturing facilities, data centers, grid infrastructure, renewable energy projects, and related utility systems. These impacts can also affect ecosystems, water basins, biodiversity, and communities, especially where data center sites intersect with water-stressed or ecologically sensitive regions.  

Companies that already have water, nature, biodiversity, or land-use goals should integrate AI-related impacts into those existing frameworks rather than create a separate silo. Practical first steps include mapping AI suppliers and infrastructure locations, requesting water- and land-use data from providers, using water-stress and biodiversity screening tools where available, and documenting where data is estimated or unavailable. Initial research emphasizes that AI-driven growth is already contributing to water stress, land conversion, and upstream hardware impacts, while standardized measurement is still emerging.  

How can companies use AI to advance sustainability innovation and workforce capabilities?

AI can support sustainability innovation when it improves outcomes without creating unmanaged environmental or social costs. The goal should not be to add more tools or workflows, but to reduce friction, improve decision-making, and help sustainability teams do more effective work with constrained resources. Potential use cases include resource optimization, climate modeling, energy management, logistics improvements, sustainability data analysis, reporting support, and internal knowledge management. However, we recommend that companies evaluate where AI is the right tool for the problem, whether the expected benefit outweighs the environmental footprint, and if the use case can be governed responsibly. This may include assessing responsible use risks and limitations (e.g., data quality, bias, hallucinations), clearly defining the intended sustainability objective, estimating expected efficiency gains, and determining how success will be measured before scaling.

How does AI adoption affect corporate climate, decarbonization, and net-zero goals? 

AI adoption can change a company’s business-as-usual emissions trajectory, especially where AI becomes embedded in products, services, operations, and infrastructure strategy. Many climate targets were set before today’s pace of AI adoption, and older baselines may not fully reflect AI-driven growth in electricity demand, cloud services, hardware needs, or customer product use. This does not justify weakening climate commitments. Instead, companies should revisit assumptions underlying their climate targets and net-zero plans, update growth scenarios, and integrate AI-related emissions into decarbonization pathways, procurement decisions, product planning, and capital allocation.

This right approach is to preserve long-term target integrity while evolving the levers used to meet those targets. Over time, better supplier data, product carbons footprints, and Scope 3 data systems should improve business’ capabilities to track reductions across the value chain. Early evidence suggests that AI is scaling faster than existing environmental accounting practices, leaving important gaps in the measurement of lifecycle impacts and value chain emissions, particularly across Scope 3 categories and outsourced digital infrastructure. This may limit companies' ability to accurately assess how AI adoption affects their emissions trajectory and identify the most effective decarbonization interventions.

How should companies respond to customer requests for AI product-level emissions data?

We recommend that companies respond transparently and avoid false precision. Customer demand for AI product-level emissions data is moving faster than the construction of data infrastructure needed to provide fully comparable, lifecycle-based metrics. A credible response should explain what is currently measured, what is estimated, what is excluded, which suppliers or cloud providers are covered, and when the methodology will be improved. Cloud dashboards and provider tools can help, but they vary in coverage, granularity, and methodological assumptions. Companies should avoid presenting narrow metrics, such as energy per token, as a complete product footprint. Instead, they should move towards lifecycle framing and supplier-specific product carbon footprint approaches that connect customer requests with procurement asks, product design decisions and annual improvement cycles. BSR members indicate strong demand from their customers for granular AI product-level emissions data and better access to high-quality data from AI providers.

What standards, methodologies, and reporting frameworks should companies use for AI-related environmental reporting?

We recommend companies start with established lifecycle, ICT, software, and corporate GHG accounting principles, while recognizing that AI-specific methodologies are still developing. The practical task is to build a defensible, transparent process: define the activity being measured, set boundaries, and improve estimates over time. Emerging AI-specific tools and methods, including energy efficiency scores and software carbon methods, can support screening and internal decision-making, but the field remains fragmented. Companies can work with solutions providers and technical partners that can align measurement to recognized standards rather than rely on opaque proprietary estimates. The current state of play suggests there is not yet a widely adopted standardized approach to measuring AI-related energy, water, and lifecycle emissions, which limits comparability and credible disclosure.

What procurement questions should companies ask AI suppliers, cloud providers, and data center providers?

The right procurement questions depend on the AI use cases, such as general cloud services, externally sourced AI models (or AI model APIs), customized models, embedded product features, enterprise AI tools, or dedicated hosting. At a minimum, companies should ask suppliers to provide emissions and energy data attributable to the relevant service; explain the methodology used, including system boundaries, assumptions, allocation approaches, and exclusions; disclose data center geography, electricity sourcing and renewable energy approach, and water use and cooling practices; and, most critically, describe plans to reduce impacts over time.

Companies should also ask whether suppliers can provide product or service-level carbon data, whether they have a GHG inventory and climate targets, and how their services support the buyers Scope 3 reduction efforts. Screening tools, such as the AI Energy Score, may help procurement teams compare models, but they should sit within a broader lifecycle and responsible sourcing framework. Procurement can also be an important lever for transparency. Through BSR's Environmental Impacts of AI Working Group, member companies are exploring how shared supplier expectations and collective buyer influence can drive improved disclosure of energy, water, carbon, and community impacts across the AI value chain.

How can companies mitigate the environmental impacts of AI adoption?

We recommend that companies treat mitigation of AI’s impacts as part of responsible AI governance, not as a standalone sustainability exercise. At the organizational level, this means establishing policies, governance structures, and review processes, among other measures, that consider environmental and social impacts before AI tools are procured, deployed, or embedded in products. Levers to mitigate environmental impacts may include reusing or fine-tuning existing models where appropriate, right-sizing models, optimizing model revision, using efficient infrastructure, adopting carbon-intelligent scheduling and monitoring, and selecting lower-impact hosting environments or low-value AI use. As environmental and social risks rise across AI infrastructure and the broader value chain, organizations incorporating these considerations into governance and deployment will be better served to build trust and future-proof investments.

How can companies govern and upskill employees for responsible, efficient AI use? 

Employee upskilling should go beyond prompt engineering. We recommend that companies train employees to understand when AI is appropriate, which tools or models are suitable for different tasks, how to avoid unnecessary or repetitive high-compute use, and how to recognize workflows where AI may introduce errors, bottlenecks, or inefficiencies. Sustainability teams should work with responsible AI teams/committees, IT, procurement, and product teams to create practical decision trees, model use guidance, and feedback loops that can be tracked over time.

Employee communications should also contextualize AI emissions at the company or product level rather than rely on abstract per-prompt comparisons, which can feel disconnected from business decision. A useful model is to pair general education on responsible and efficient AI use with product- or function-specific learning, so employees understand both the environmental implications of AI and the ways AI can improve their work responsibly. Across BSR members, employee upskilling, responsible AI use, and product-based learning are emerging priorities, along with seeking ways to respond to employee questions about prompt guidance, appropriate AI use, and company policies governing AI adoption. 

How can collaboration help companies address AI’s environmental impacts?

Conclusion 

Companies do not need to wait for perfect standards to begin managing AI’s environmental impacts. The near-term priority is to build repeatable processes by mapping AI use, defining boundaries, improving supplier data, integrating AI into climate and nature strategies, and using procurement and governance to shape better outcomes. BSR members can use collaborative spaces to align expectations, reduce duplications, and accelerate practical measurement and mitigation across the AI value chain.