The Debate Around AI’s Environmental Impact
Artificial intelligence (AI) has become impossible to ignore. From breakthrough models to the wave of new applications reshaping business and society, AI is everywhere. But alongside the optimism sits a harder question: what does AI mean for the environment?
The reality is complicated. AI is both a consumer of vast resources and a potential enabler of climate solutions. It is reshaping how we approach energy, industry, agriculture and research and innovation itself, while simultaneously driving demand for energy-hungry data centres and resource-intensive hardware. On balance, can it deliver a net positive impact on the environment?
AI, Machine Learning and Generative models explained
AI is not new. The concept dates back to the 1950s, and many of today’s algorithms, such as those used in machine learning, have developed gradually over decades.
It is common for different technologies to be grouped together under the umbrella of “AI”. Yet to understand AI’s sustainability, distinguishing between them is important, both to appreciate their capabilities and to assess their environmental footprint.
The image illustrates the hierarchy of AI systems:
- Artificial Intelligence (AI): The broad field of developing systems that can perform tasks typically requiring human intelligence, from recognising patterns to making decisions.
- Machine Learning (ML): A subset of AI where algorithms identify patterns in data and use these patterns to make predictions or decisions. Instead of being explicitly programmed for each task, ML systems improve as they are exposed to more data.
- Deep Learning: A branch of ML that uses multi-layered neural networks to handle highly complex problems such as image recognition, natural language processing and speech.
- Generative AI: Models trained not only to analyse data but to create new content, text, images, video or code, based on patterns they have learned.
- Large Language Models (LLMs): A type of generative AI built on transformer architectures, capable of producing coherent, human-like responses at scale. Recent examples include GPT-4 and Gemini, which have driven much of the recent hype.
The deeper down this stack you go, the more computationally intensive, and therefore environmentally demanding, the systems become.
To understand the reason for the “AI wave” we have to understand the advancements that have brought AI to what it is today. The publication of the transformer architecture in 2017, and the launch of ChatGPT in 2022, marked a step change in capability and accessibility. Suddenly, AI was no longer a niche field but a mainstream technology capable of producing fluid, useful outputs. The result has been an explosion of investment from big tech firms and venture capital, all racing to dominate the next great platform shift.
This excitement has bundled very different technologies, from narrow machine learning models to large language models, into a single narrative. That bundling is obscuring the fact that not all AI is equal, and neither are their environmental footprints.
Environmental Costs of AI and Climate Change: Data Centres, Energy and Water Use
Data centres are central to discussions about AI’s carbon footprint and energy demand, and the growth of data centres is intrinsically linked to the growth of AI. These facilities, which house the servers and GPUs used to train and run AI models, have seen investment almost double since 2022, reaching half a trillion dollars in 20242. The International Energy Agency projects that AI could drive a 160% increase in global data centre power demand by 20302.
The consequences are significant. Data centres already account for around 1.5% of global electricity use, a figure expected to more than double by the end of the decade2.
The Accenture analysis above3 shows how steep the growth of global data centre energy demand could be.
The wider environmental costs include:
- Electricity use: data centres must run constantly so often draw on fossil-fuelled grids rather than intermittent renewable energy
- Construction emissions: from steel, cement and supply chains
- Water consumption: vast amounts of water are required for cooling, often in regions already facing water scarcity
- Hardware impacts: mining, manufacturing and disposal of specialised equipment such as GPUs create additional carbon and e-waste challenges
Together, these factors make the environmental cost of AI infrastructure clear. In short, the AI environmental impact is not limited to electricity demand, but extends to water, construction, and resources too.
Even companies with strong sustainability commitments are struggling to reconcile ambitions with reality. Google has reported a 48% rise in emissions over the past five years, largely attributed to the energy demands of its expanding AI infrastructure4.
Positive Applications of AI in Sustainability and Climate Solutions
On the other hand, AI for sustainability is already being used to reduce environmental impacts across a wide range of sectors.
In energy, AI helps manage increasingly complex grids, forecasting demand and balancing intermittent renewables. In buildings, intelligent control systems optimise heating, cooling and lighting to cut waste. Industry is AI for enhanced process control in manufacturing to improve energy efficiency and enable predictive maintenance, while transport planners apply AI to optimise travel, reduce congestion and fuel use. Agriculture is benefitting from precision farming techniques that minimise fertiliser and water use, and climate scientists are deploying AI to model extreme weather events with greater accuracy.
These applications of sustainable AI, and many others, are already in use today, delivering measurable improvements in efficiency and emissions reduction. They also highlight the role AI climate change tools can play in sectors as diverse as energy, transport, agriculture and waste management.
How AI Speeds Up Innovation in Clean Technologies
Perhaps the most exciting potential lies in how AI accelerates innovation itself. Scientific discovery has traditionally relied on laborious cycles of trial and error. AI can compress these cycles dramatically. Some examples include2,5:
- Material science: less than 0.01% of potential next-generation solar PV compounds have been tested. AI can model the rest, narrowing the field for laboratory validation and prototyping. Similar approaches are already helping researchers screen new battery chemistries or carbon capture molecules.
- Sustainable agriculture: AI can be used for faster discovery of new active ingredients and modes of action for crop protection and seed resilience selection
- Renewable tech manufacturing: AI can be used to iterate through different prototype designs to speed up refinement and deployment of clean technology.
By unlocking insights from vast, complex datasets, AI offers a way to move faster towards the low-carbon solutions we urgently need.
For innovators, this acceleration can be transformative. Directing R&D capital towards the most promising pathways could mean faster time to market, quicker scaling, and ultimately a more efficient use of resources.
The Efficiency Paradox: When AI Increases Emissions
But efficiency alone does not guarantee sustainability. AI’s power lies in its ability to optimise, and that ability can be applied anywhere. The same algorithms that help balance a renewable-heavy grid can also make oil drilling more profitable or fast fashion marketing more persuasive.
This is the efficiency paradox. By lowering costs and raising productivity, AI risks increasing overall consumption even as it reduces emissions per unit of output. Without careful governance and direction, AI could just as easily accelerate unsustainable practices as it could climate solutions.
Funding Trends: Is AI Crowding Out Climate Innovation?
The investment landscape reflects these tensions. While funding for climate tech has declined for three consecutive years, AI investment has surged6. A third of investors report focusing on opportunities at the intersection of AI and climate, but others acknowledge concerns that the enthusiasm for AI is crowding out capital for hardware-intensive solutions7 such as hydrogen, nuclear or advanced materials.
AI startups are typically software-based, scale quickly, and offer the prospect of faster returns. By contrast, hard-tech ventures require significant capital, longer development cycles and carry higher risks. Yet these are the technologies needed for deep, system-level decarbonisation.
Can AI Really Have a Net Positive Environmental Impact?
So can AI have a net positive impact on the environment? The answer is not straightforward. Its footprint, in energy, water and materials, is significant and growing. At the same time, it has massive potential to accelerate innovation and enable efficiency gains that lead to positive environmental outcomes.
On balance, AI can deliver a positive outcome if we meet three conditions:
- Applying the right models to the right problems, focusing resource-intensive AI on challenges where the benefits justify the cost
- By aligning investment so that AI complements rather than displaces cleantech/climatetech
- By greening the infrastructure that underpins AI, from data centres to hardware supply chains. We will explore this last point in more depth in an upcoming article on how innovations can make AI infrastructure more sustainable.
What is clear for now is that AI’s climate impact will be decided both by the technology itself and by how it is applied. Used wisely, it has the potential to be a valuable tool in the transition to net zero. Directing sustainable AI towards the toughest climate challenges will determine whether its overall impact is net positive.
Key takeaways
- AI’s climate impact is double-edged: it consumes significant energy, water and materials, yet also enables efficiency gains and innovation that can reduce emissions.
- Not all AI is equal: large language models are far more resource-intensive than traditional machine learning, which often delivers impact at a fraction of the footprint.
- Data centres are the bottleneck: AI is driving a surge in electricity demand, construction, water use and e-waste, creating growing pressure on sustainability commitments.
- Real-world benefits are emerging: from smart grids to precision agriculture, AI is already cutting emissions and resource use in multiple sectors.
- AI accelerates discovery: by narrowing the search for new materials, batteries and carbon capture molecules, AI can speed up the development of key climate technologies.
- Efficiency is not enough: without careful application, AI can just as easily optimise oil drilling or fast fashion as it can decarbonisation.
- Capital is shifting: investors are favouring AI software over capital-intensive hard tech, raising concerns about whether critical climate technologies are being crowded out.
- The balance will depend on choices: AI’s net climate impact will be shaped both by the technology itself and by how deliberately it is applied to the toughest decarbonisation challenges.
FAQs
Why does AI use so much energy?
Training and running advanced AI models involves billions of calculations. This requires powerful hardware, usually clustered in large data centres that run continuously and need substantial cooling.
Why is it that large language models have a greater environmental impact than traditional machine learning?
Because LLMs are trained on vastly larger datasets and require billions of parameters, LLMs and generative AI need far more computing power, electricity and cooling. Traditional machine learning models are smaller, task-specific, and therefore much less resource-intensive.
What is sustainable AI?
Sustainable AI refers to designing, training and deploying AI systems in ways that minimise their carbon footprint, for example by using renewable-powered data centres, more efficient algorithms, or recycling hardware.
How can companies reduce the footprint of their AI use?
By selecting smaller, more efficient models where possible, sourcing renewable-powered data centre services and limiting non-essential use cases.
What innovations could make AI more sustainable in the future?
Promising areas include liquid cooling systems, renewable-powered data centres, semiconductor efficiency improvements, and circular approaches to GPU manufacturing and recycling. Keep an eye out for an upcoming article on this topic.
References
- Tiny Technical Tutorials: AI vs. Machine Learning vs. Deep Learning vs. Generative AI: What’s the Difference? – https://open.spotify.com/episode/31g38pEgQDboZTELzIqPWP?si=gmIlFwe5TzC9RJygMJUj4A&context=spotify%3Aplaylist%3A37i9dQZF1FgnTBfUlzkeKt
- International Energy Agency (IEA): Energy Demand from AI – https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
- Accenture: Powering Sustainable AI – https://www.accenture.com/content/dam/accenture/final/corporate/corporate-initiatives/sustainability/document/Powering-Sustainable-AI.pdf#zoom=40
- The Guardian: Google’s emissions climb nearly 50% in five years due to AI energy demand – https://www.theguardian.com/technology/article/2024/jul/02/google-ai-emissions
- World Economic Forum (WFE): Post breakthrough: How AI can lift climate research out of the lab and into the real world – https://www.weforum.org/stories/2024/05/ai-lift-climate-research-out-lab-and-real-world/
- Pitchbook|NVCA: Venture Monitor 2024 – https://nvca.org/wp-content/uploads/2025/01/Q4-2024-PitchBook-NVCA-Venture-Monitor.pdf
- Sightline Climate: Our 2025 climate tech investor pulse check https://www.ctvc.co/2025-climate-tech-investor-pulse-check/
Written by