Artificial Intelligence Becomes a Major Consumer of Energy and Water: How the Growth of Neural Networks Affects the Climate and the Risks and Opportunities This Presents for Investors and the Global Economy
The Surge in AI and its Appetite for Energy
The demand for computational power for AI has soared in recent years. Since the launch of public neural networks like ChatGPT at the end of 2022, businesses worldwide have accelerated the adoption of AI models, necessitating vast amounts of data processing. Industry estimates suggest that by 2024, AI may account for around 15–20% of the total energy consumption of data centres globally. The power required for AI systems could reach 23 GW by 2025 – comparable to the total electricity consumption of a country like the United Kingdom. For context, this figure exceeds the energy consumption of the entire Bitcoin mining network, indicating that AI has become one of the most energy-intensive forms of computing.
This exponential dynamic is driven by massive investments from technology companies in infrastructure: new data centres are opened almost weekly, and specialised chips for machine learning are being produced every few months. The expansion of such infrastructure directly leads to an increase in electricity consumption necessary to power and cool thousands of servers supporting contemporary neural networks.
Emissions on the Scale of Megacities
Such high energy consumption inevitably results in significant greenhouse gas emissions if part of the energy is derived from fossil fuels. According to recent research, AI could be responsible for 32–80 million metric tonnes of CO2 emissions annually by 2025. This effectively places the "carbon footprint" of AI on par with that of an entire city: for example, New York generates around 50 million tonnes of CO2 per year. For the first time, a technology previously seen as purely digital demonstrates a climate impact scale comparable to that of major industrial sectors.
It is important to highlight that these estimates are considered conservative. They primarily account for emissions from electricity generation to operate the servers, while the complete lifecycle of AI – from the production of equipment (servers, chips) to disposal – generates an additional carbon footprint. Should the AI boom continue at its current pace, the volume of related emissions will rise sharply. This complicates global efforts to reduce greenhouse gases and places a challenge on technology companies – how to integrate the explosive growth of AI into their commitments to achieve carbon neutrality.
The Water Footprint of Neural Networks
Another hidden resource 'appetite' of AI is water. Data centres consume enormous amounts of water for cooling servers and equipment: evaporative cooling and air conditioning cannot function without water resources. In addition to direct consumption, significant volumes of water are indirectly needed – at power plants for cooling turbines and reactors in the generation of the very electricity consumed by computational clusters. Experts estimate that AI systems alone could consume between 312 and 765 billion litres of water by 2025. This is on par with the total volume of bottled water consumed by humanity in a year. Thus, neural networks generate a colossal water footprint that has, until recently, gone largely unnoticed by the public.
Official estimates often fail to provide a complete picture. For instance, the International Energy Agency cited a figure of approximately 560 billion litres of water used by all data centres worldwide in 2023; however, this statistic did not include water used at power plants. The actual water footprint of AI may be several times higher than formal estimates. Industry giants are currently reluctant to disclose details: in a recent report on its AI system, Google explicitly stated that it does not account for water consumption at third-party power plants. Such an approach is under scrutiny, as a significant portion of water is consumed to meet AI's electricity needs.
The scale of water consumption is already raising concerns in several regions. In arid areas of the US and Europe, communities are opposing the construction of new data centres, fearing they will draw scarce water from local sources. Corporations themselves are noting an increase in the 'thirst' of their server farms: for instance, Microsoft reported that global water consumption by its data centres surged by 34% (to 6.4 billion litres) in 2022, largely due to increased loads associated with training AI models. These facts underscore that water management is rapidly coming to the forefront in assessing the environmental risks of digital infrastructure.
The Opacity of Tech Giants
Paradoxically, despite the scale of their impact, data on energy and water consumption by AI is largely unavailable to the public. Major technology companies (Big Tech) in their sustainability reports typically present aggregated figures on emissions and resource usage without separately highlighting the share attributed to AI. Detailed information on the operations of data centres – such as how much energy or water is used specifically for neural network computations – often remains internal to the companies. There is virtually no information on 'indirect' consumption, such as water used in the production of electricity for data centre needs.
Consequently, researchers and analysts have to act like detectives, piecing together the picture from fragmented data: snippets from corporate presentations, estimates of the number of AI server chips sold, data from energy companies, and other indirect indicators. This lack of transparency complicates understanding the full scale of AI's environmental footprint. Experts are calling for stringent reporting standards: companies should disclose energy consumption and water usage of their data centres, broken down by key domains, including AI. Such transparency would enable society and investors to objectively assess the impact of new technologies, and would encourage the industry to seek ways to reduce its environmental burden.
Pervasive Environmental Risks
If current trends continue, the growing 'appetite' for AI could exacerbate existing environmental issues. Additional tens of millions of tonnes of greenhouse gas emissions annually would complicate meeting the goals of the Paris Climate Agreement. The consumption of hundreds of billions of litres of fresh water will occur against the backdrop of a global shortage of water resources, which is projected to reach 56% by 2030. In other words, without sustainable development measures, the expansion of AI risks conflicting with the planet's environmental limits.
If no changes are made, such trends could lead to the following negative consequences:
- Accelerated global warming due to increased greenhouse gas emissions.
- Worsening fresh water shortages in already arid regions.
- Increased strain on energy systems and socio-environmental conflicts over limited resources.
Local communities and governments are already beginning to respond to these challenges. In some countries, restrictions are being implemented on the construction of 'energy-hungry' data centres, requiring the use of water recycling systems or renewable energy purchasing. Experts note that without significant changes, the AI industry risks transitioning from a purely digital domain into a source of tangible environmental crises – from droughts to the derailment of climate plans.
Investor Perspective: The ESG Factor
Environmental aspects of the rapid development of AI are becoming increasingly important for investors as well. In an era when ESG (Environmental, Social, and Governance) principles take centre stage, the carbon and water footprints of technologies directly influence company valuations. Investors are asking whether the 'green' shift in policy will result in increased costs for companies betting on AI. For instance, tightening carbon regulations or introducing fees for water use could raise expenses for companies whose neural network services consume large amounts of energy and water.
Conversely, companies investing now in mitigating the environmental impact of AI may gain a competitive advantage. Transitioning data centres to renewable energy, improving chips and software for enhanced energy efficiency, and implementing water reuse systems reduce risks and bolster reputation. The market highly values progress in sustainability: investors worldwide are increasingly incorporating environmental metrics into their business evaluation models. Therefore, for technology leaders, the question is urgent: how to continue to scale AI capabilities while meeting societal expectations for sustainability? Those who find a balance between innovation and responsible stewardship of the planet will win in the long term – both in terms of reputation and business value.
The Path to Sustainable AI
Despite the scale of the challenge, the industry has opportunities to steer the growth of AI towards sustainable development. Global technology companies and researchers are already working on solutions capable of reducing AI's environmental footprint without hindering innovation. Key strategies include:
- Improving energy efficiency of models and equipment. Developing optimised algorithms and specialised chips (ASICs, TPUs, etc.) that perform machine learning tasks with lower energy consumption.
- Transitioning to clean energy sources. Harnessing electricity from renewable resources (solar, wind, hydro, and nuclear power) to power data centres, thereby eliminating carbon emissions from AI operations. Many IT giants are already signing 'green' contracts to purchase clean energy for their needs.
- Reducing and recycling water consumption. Implementing new cooling systems (liquid and immersion cooling) that require significantly less water, as well as reusing technical water. Selecting locations for data centres with consideration of water availability: favouring regions with cool climates or sufficient water resources. Research shows that careful selection of location and cooling technologies can reduce the water and carbon footprint of a data centre by 70–85%.
- Transparency and accountability. Instituting mandatory monitoring and reporting of energy consumption and water use by AI infrastructure. Public accountability encourages companies to manage resources more efficiently and allows investors to track progress in reducing ecosystem burdens.
- Utilising AI for resource management. Paradoxically, AI itself can help solve this problem. Machine learning algorithms are already being applied to optimise cooling in data centres, predict loads, and allocate tasks to minimise peak demands on networks and enhance the efficiency of server utilisation.
The next few years will be crucial in integrating sustainability principles into the core of the rapidly growing AI sector. The industry stands at a crossroads: either continue on inertia, risking conflict with environmental boundaries, or transform the problem into an impetus for new technologies and business models. If transparency, innovation, and responsible resource management become integral to AI strategies, the 'digital mind' can evolve hand in hand with consideration for the planet. Such a balance is what investors and society at large expect from this new technological era.
