Artificial intelligence is transforming industries at speed. From generative AI and machine learning models to automation tools and data driven decision making, AI systems are now embedded across the global economy.
But while AI often feels intangible, the infrastructure powering it is not.
A recent peer reviewed report published in Patterns by Cell Press examines the environmental impact of artificial intelligence, focusing specifically on carbon emissions and water consumption linked to AI driven data centres. The findings highlight the growing physical footprint of AI technology and raise important questions about sustainability, transparency, and long term impact.
This article explores the key findings of the report, how AI contributes to carbon and water use, and why these issues matter for companies building and deploying AI at scale.
How Much Carbon Does AI Produce?
According to the report, artificial intelligence systems could be responsible for between 32.6 million and 79.7 million tonnes of CO₂ emissions per year by the end of 2025.
These emissions are generated primarily through electricity consumption. AI models require significant computational power during both training and deployment. This includes:
- Training large language models
- Running inference for generative AI tools
- Powering AI workloads in cloud data centres
- Supporting machine learning applications across enterprises
The study focuses on operational emissions, meaning carbon produced while AI systems are running. It does not fully account for emissions associated with manufacturing hardware, building data centres, or producing semiconductors. As a result, the total lifecycle carbon footprint of AI may be even higher.
Why AI Data Centres Consume So Much Energy
AI data centres operate continuously and at high intensity. Unlike traditional computing tasks, AI workloads often require specialised hardware such as GPUs and accelerators that consume more electricity.
The report estimates that AI related electricity demand could reach up to 23 gigawatts globally by 2025, equivalent to the output of multiple large power plants.
Where that electricity comes from is critical. AI systems powered by fossil fuel heavy energy grids produce significantly more carbon emissions than those supported by renewable energy sources. This means the environmental impact of AI varies greatly by region and energy mix.
AI Water Consumption Is Often Underestimated
One of the most important findings in the report relates to water usage, an environmental cost that is frequently overlooked in discussions about AI sustainability.
The researchers estimate that AI driven data centres could consume between 312 billion and 764 billion litres of water per year.
Water consumption occurs in two main ways:
- Direct water use, primarily for cooling servers and preventing overheating
- Indirect water use, required to generate the electricity that powers AI data centres
Most public sustainability disclosures focus only on direct water use. The report shows that indirect water consumption can be significantly larger and is rarely reported. This means the true water footprint of artificial intelligence is likely being underestimated across the industry.
In regions facing water scarcity, this raises serious concerns about the long term viability of large scale AI infrastructure.
Why AI Environmental Estimates Vary So Widely
The report presents wide ranges for both carbon emissions and water consumption. This reflects a lack of transparency rather than uncertainty in the underlying science.
Most technology companies do not break down environmental metrics by workload. AI usage is grouped together with cloud computing, search, streaming, and other digital services. As a result, researchers must rely on modelling and assumptions to isolate AI’s specific impact.
The authors highlight the need for:
- Disaggregated reporting on AI energy use
- Clear carbon and water metrics tied directly to AI workloads
- Greater transparency at data centre and regional levels
Without this data, it is difficult for policymakers, businesses, and the public to accurately assess the environmental cost of artificial intelligence.
Is AI Environmentally Unsustainable?
On an individual level, a single AI query or prompt uses a relatively small amount of energy and water. The challenge lies in scale.
As AI adoption accelerates across industries, billions of AI driven interactions occur daily. This includes enterprise automation, customer support chatbots, AI powered analytics, and generative content tools. Training increasingly large models adds further strain to global infrastructure.
The report makes it clear that AI’s environmental impact is not about individual usage, but about cumulative global demand.
Why AI Sustainability Matters for the Future
Artificial intelligence is expected to play a central role in economic growth, innovation, and digital transformation. Understanding its environmental impact is essential to ensuring that growth is sustainable.
The report calls for:
- Improved transparency in AI environmental reporting
- Investment in renewable energy for AI data centres
- Adoption of water efficient cooling technologies
- Policy frameworks that account for digital infrastructure impacts
Rather than rejecting AI, the research encourages a more responsible approach to how AI systems are built, deployed, and powered.
Key Takeaway
Artificial intelligence is not just a digital technology. It is a physical system with real carbon emissions and water consumption. As AI continues to scale, its environmental footprint will become increasingly important.
This report provides a valuable foundation for understanding the true cost of AI and highlights the urgent need for clearer data, better reporting standards, and more sustainable infrastructure.
