In the first part of this series, we explored how artificial intelligence (AI) can improve energy efficiency, support renewable energy integration, optimize industrial production, and contribute to more sustainable agriculture. These applications demonstrate AI’s considerable potential to support the objectives of the European Union’s Fit for 55 package and the broader European Green Deal.
However, AI is not an environmentally neutral technology. Behind every AI-generated response, predictive model, or optimization algorithm lies an extensive digital infrastructure that requires electricity, water, critical raw materials, and increasingly powerful computing hardware.
As AI adoption accelerates across industries, policymakers and researchers are asking an important question: Can AI truly support sustainability if its own environmental footprint continues to grow?
The answer is complex. AI has the potential to reduce emissions in many sectors, but its benefits must be weighed against the resources required to develop, train, deploy, and operate increasingly sophisticated models.
AI’s growing demand for electricity
The rapid expansion of artificial intelligence is significantly increasing global electricity demand.
Unlike conventional software, modern AI models require vast computational resources for both training and inference, the process of generating responses after deployment. Large language models, image-generation systems, and industrial AI applications all rely on thousands of high-performance graphics processing units (GPUs) operating simultaneously in hyperscale data centres.
According to the International Energy Agency (IEA), global electricity consumption by data centres could exceed 1,000 terawatt-hours (TWh) annually by 2030, driven largely by AI and cloud computing. This would represent more electricity than many industrialized countries consume in an entire year.
Today, data centres already account for approximately 1-1.5% of global electricity demand, and their share is expected to grow rapidly over the coming decade.
The climate impact of AI therefore depends not only on how much electricity it consumes, but also on how that electricity is generated. Data centres powered primarily by renewable energy have a substantially lower carbon footprint than those relying on fossil fuels.
Data centres consume more than electricity
Electricity is only one component of AI’s environmental footprint.
High-performance processors generate enormous amounts of heat while performing billions of calculations every second. Without continuous cooling, servers would quickly overheat, reducing both performance and equipment lifespan.
Modern data centres therefore require sophisticated cooling systems, including:
- air cooling,
- liquid cooling,
- evaporative cooling,
- chilled water systems,
- heat recovery technologies.
Cooling infrastructure represents a substantial share of a data centre’s total energy demand and has become one of the industry’s greatest engineering challenges.
As AI hardware becomes more powerful, managing the heat generated by densely packed GPU clusters is becoming increasingly difficult. AIRSYS ANSI
Water: The often overlooked resource
Perhaps the least visible aspect of AI’s environmental footprint is water consumption.
Many large data centres rely on evaporative cooling systems that consume significant volumes of freshwater to dissipate heat.
Research conducted by the University of California, Riverside, together with the University of Texas at Arlington, estimated that generating a series of interactions with a large language model can indirectly consume approximately 500 millilitres of water, depending on server location, cooling technology, and electricity generation. ACM
On a global scale, the numbers become substantial.
Microsoft reported a 34% increase in water consumption between 2021 and 2022, largely driven by expanding AI infrastructure. Google also reported increasing freshwater use as its data centre operations expanded to support AI services.
These figures do not imply that AI itself “uses” water directly. Rather, water is consumed by the cooling systems required to maintain safe operating temperatures within large-scale computing facilities.
In regions already experiencing water scarcity, this has become an important topic in discussions about sustainable digital infrastructure.
The hidden footprint of AI hardware
The environmental impact of AI begins long before a model is trained.
Modern AI depends on highly specialized processors manufactured through one of the world’s most resource-intensive industrial processes.
Producing advanced semiconductors requires:
- enormous quantities of electricity,
- ultra-pure water,
- critical minerals,
- rare earth elements,
- and highly complex chemical manufacturing processes.
Can efficiency offset growing demand?
One of the central debates surrounding AI concerns the so-called rebound effect.
Artificial intelligence often improves efficiency by reducing energy use, optimizing logistics, or streamlining manufacturing processes. However, increased efficiency can also reduce operating costs, encouraging greater overall consumption. Medium
For example:
- more efficient computing may increase demand for AI-powered services
- optimized logistics can stimulate higher transport volumes
- cheaper digital services may encourage wider adoption
As a result, some of the environmental gains achieved through efficiency improvements may be offset by increased economic activity.
Researchers therefore emphasize that technological efficiency alone cannot guarantee lower emissions without supportive policy measures.
AI, regulation, and the Fit for 55 agenda
The European Union increasingly recognizes both the opportunities and the challenges associated with artificial intelligence.
While the AI Act primarily focuses on safety, transparency, and fundamental rights, broader European policies, including the European Green Deal, Fit for 55, CSRD, and the Energy Efficiency Directive, encourage organizations to consider the environmental impacts of digital technologies.
For businesses, this means that future AI strategies are likely to be evaluated not only according to their technical capabilities but also their energy efficiency, resource consumption, and contribution to corporate sustainability objectives. European Commission
As Europe moves toward climate neutrality, AI will likely become part of a broader conversation about responsible digitalization rather than simply technological innovation.
Conclusion
Artificial intelligence presents both significant opportunities and important environmental challenges.
On one hand, AI can help optimize energy systems, improve industrial efficiency, support renewable energy integration, and accelerate climate adaptation. On the other, it depends on an expanding digital ecosystem that consumes electricity, water, raw materials, and increasingly sophisticated hardware.
From the perspective of the Fit for 55 strategy, the objective should not be to maximize AI deployment at any cost, nor to reject the technology because of its environmental footprint. Instead, the challenge is to ensure that AI delivers greater environmental benefits than the resources it consumes.
Achieving this balance will require continued improvements in energy-efficient hardware, renewable-powered data centres, sustainable cooling technologies, responsible supply chains, and circular management of digital infrastructure.
Ultimately, AI should be viewed neither as a climate solution nor as a climate problem in itself. Like many transformative technologies, its long-term sustainability will depend on how it is designed, powered, regulated, and applied.
