Part 2 of our 5-part series.

AI, large language models, and corporate net-zero targets.

Sustainability enabler or a net-zero target risk?

As businesses accelerate their digital transformation, many are developing propriety Large Language Models (LLMs) trained on company data to enhance operational efficiency. The benefits are clear—streamlined processes, automated knowledge management, and more informed decision-making. However, as AI becomes increasingly embedded in corporate functions, a critical question arises: how does AI’s energy consumption align with corporate net-zero commitments? 

AI in business operations: A sustainability opportunity 

The adoption of LLMs in corporate environments offers substantial efficiency gains. AI-powered knowledge management tools enable businesses to automate routine queries in legal, HR, and compliance functions, reducing administrative overheads. AI also enhances decision-making by analysing corporate emissions data, financial statements, and regulatory requirements to identify sustainability risks and optimise resource allocation. Additionally, AI-driven automation within IT infrastructure can help reduce redundancy, optimise cloud workloads, and increase energy efficiency across digital operations. By integrating LLMs into internal processes, businesses can boost productivity while indirectly reducing emissions. However, AI’s role as a sustainability enabler must be carefully weighed against its increasing energy demands. 

The carbon cost of training AI 

AI’s rising energy demands pose a significant challenge to corporate net-zero strategies, particularly through Scope 2 emissions for data centre operators and Scope 3 emissions for businesses relying on cloud-based AI services. With data centres already consuming 1% to 2% of global electricity – a figure projected to rise to 21% by 2030¹ – the sustainability risks associated with AI adoption are escalating. 

For data centre owners and cloud service providers, AI represents a growing scope 2 emissions challenge. Operating large-scale servers and cooling systems requires vast amounts of electricity, much of which is still sourced from fossil fuels. A 2023 MIT study found that training a single AI model from scratch can consume up to 1,000 megawatt-hours (MWh) of energy – enough to power a small town for a year. Another study found that training just one large AI model can emit over 300,000 kgCO₂e, equivalent to the lifetime emissions of five cars. Unless data centres transition to renewable energy-powered infrastructure, their emissions will continue to rise. 

For corporates leveraging AI through cloud services, data centre emissions fall within scope 3 under purchased goods and services or leased assets, depending on how cloud computing is procured. Companies renting dedicated cloud infrastructure may classify these emissions under Leased Assets, whereas businesses using shared cloud computing resources will likely categorise them under Purchased Goods and Services. However, many organisations fail to track and report these emissions, leading to an underestimation of AI’s carbon footprint in corporate sustainability disclosures. 

Measuring AI’s Carbon footprint 

Quantifying AI’s carbon footprint is complex, but emerging methodologies are helping businesses assess and manage AI-related emissions: 

1.  Energy Consumption Monitoring – Cloud providers such as Google Cloud, Microsoft Azure, and AWS offer dashboards that track the electricity usage of AI workloads, providing insights into their environmental impact.
2. Life Cycle Assessment (LCA) of AI Models – A comprehensive LCA assesses emissions from hardware sourcing (e.g., GPUs and data centre infrastructure) to operational energy use and eventual decommissioning.
3. Cloud Provider Sustainability Reports – Businesses should engage with cloud providers that disclose carbon intensity metrics. For instance, Google Cloud’s Carbon Footprint tool allows companies to track emissions from their cloud-based AI workloads.
4. Integration with Corporate Carbon Accounting – AI-related emissions should be incorporated into Scope 2 and Scope 3 reporting in alignment with the Greenhouse Gas (GHG) Protocol, ensuring transparency in sustainability disclosures.

Balancing AI innovation with sustainability goals 

To mitigate AI’s environmental impact while maximising efficiency, businesses must embed sustainability into their AI adoption strategies. Instead of training large models from scratch, companies can fine-tune pre-trained models, significantly reducing computational intensity. Additionally, deploying smaller, task-specific AI models for internal applications can lower energy consumption without compromising effectiveness. Further strategies for aligning with net-zero targets include choosing cloud providers with strong sustainability commitments, such as those implementing carbon-aware load balancing and integrating AI emissions data into corporate sustainability decision-making to ensure AI-driven efficiencies align with net-zero ambitions. 

AI as a sustainability enabler or risk? 

AI has the potential to drive business efficiencies and support sustainability efforts, but its energy demands pose a significant challenge. As AI adoption grows, companies must balance technological innovation with environmental responsibility. Those that proactively measure, manage, and mitigate AI’s carbon footprint will not only reduce sustainability risks but also position themselves as leaders in the transition to a smarter, low-carbon digital economy. 

This is the first insight in our five-part series on AI and ESG. Next, we explore AI’s hidden environmental impacts, specifically water consumption and rare earth mineral use and what this means for responsible corporate strategy. 

Author: Harry Freeman, Consultant, Simply Sustainable

1.  https://mitsloan.mit.edu/ideas-made-to-matter/ai-has-high-data-center-energy-costs-there-are-solutions