There is a conversation happening in boardrooms, on LinkedIn, and in the press and if you are investing in AI for your organisation, you are probably hearing it too. The question is straightforward: is AI bad for the environment?
The honest answer is: yes, it has a cost. But context matters enormously, and the headlines rarely provide it. As a leader, your job is to make decisions based on accurate information, not fear, and not complacency. This blog will give you exactly that.
The Real Environmental Cost of AI
AI does not run on good intentions. It runs on electricity, vast quantities of it, inside massive data centres packed with specialised hardware, operating around the clock. Those data centres also consume water, produce hardware waste, and, depending on their energy source, contribute to carbon emissions.
The numbers are significant. Global data centre electricity demand sat at approximately 415 terawatt-hours in 2024, representing around 1.5% of total global electricity use. The International Energy Agency projects this will nearly double to 945 TWh by 2030, largely driven by AI growth, comparable to Japan’s entire electricity consumption today. Carbon emissions tied to AI infrastructure are estimated to reach between 32.6 and 79.7 million tonnes of CO₂ in 2025 alone. Training a single large AI model has been estimated to emit tens of thousands of tonnes of CO₂, equivalent to thousands of cars’ annual emissions.
Water use is the less-discussed but equally real issue. Data centres use vast volumes of water to cool their servers. Microsoft acknowledged that 42% of the water it consumed in 2023 came from areas with water stress; Google reported 15% of its freshwater withdrawals came from areas with high water scarcity.
This is not nothing. Any leader who tells you AI has no environmental cost is not telling you the truth.
Now, Let's Put It in Context
Here is what the headlines rarely tell you: AI’s environmental footprint, while real and growing, is currently a fraction of the impact of activities most organisations, and most individuals, consider entirely routine.
Flying. It is estimated that a single long-haul flight from Los Angeles to London could equal the energy use of roughly 20 million AI queries. Aviation accounts for approximately 2.5% of global CO₂ emissions, more than double the current share of all data centres combined. If your leadership team flies to an annual conference, that single trip almost certainly dwarfs the carbon cost of your organisation’s AI usage for the year. (Source)
Driving. Transport globally contributes roughly a quarter of all energy-related CO₂ emissions. The cumulative impact of daily commutes and business travel vastly exceeds that of consumer AI use. A single short flight’s carbon emissions are equivalent to approximately 90,000 AI queries.
Heating buildings. Heating a typical UK home for a single day may equal around 100,000 AI queries in energy terms. For most organisations, the emissions profile of their premises will significantly outweigh their AI footprint. (Source)
Video streaming & Gaming. One hour of video streaming is comparable in energy terms to roughly 15 AI queries. Streaming platforms and YouTube together produce an estimated 50 million tonnes of CO₂ equivalents globally, comparable to, or greater than, AI’s current footprint. (Source)
Cryptocurrency mining. Bitcoin mining alone has at times consumed electricity comparable to entire countries, with data centres and crypto combined estimated at around 460 TWh annually. Unlike AI, cryptocurrency mining generates no productivity gains, scientific breakthroughs, or operational efficiencies to offset its energy cost. (Source)
Food. Producing just one pound of beef emits roughly 22 pounds of CO₂-equivalent emissions — the result of methane from cattle, vast land and water use, and deforestation. A single AI text prompt produces approximately 0.1 to 0.5 grams of CO₂e. That makes one pound of beef equivalent in emissions terms to tens of thousands of AI prompts. A business lunch with beef on the menu will almost certainly have a larger carbon footprint than a full day of AI usage across your team. (Source)
The point is not that AI is harmless. The point is that we should apply the same scrutiny to all high-energy activities, and be precise about where the real risks lie.
Where the Genuine Concern Lies
The concern is not that one employee asking an AI tool a question is catastrophic. It is not. A single ChatGPT query uses approximately 0.3 to 3 watt-hours of electricity and produces around 0.1 to 4 grams of CO₂. That is roughly comparable to running a lightbulb for a few minutes.
The legitimate concern is scale and trajectory. Billions of queries per day, ever-larger models, and rapid global data centre construction, if powered predominantly by fossil fuels, could place serious pressure on electricity grids and emissions targets over the next decade. Data centre electricity use has been growing at approximately 12% per year.
This is the distinction every leader needs to hold clearly: individual use is not the problem; unchecked industrial-scale expansion powered by fossil fuels is the risk.
AI Can Also Be Part of the Solution
A balanced assessment requires acknowledging something important: AI is not only a consumer of energy — it is increasingly a tool for reducing it.
Google’s DeepMind reported using AI to reduce data centre cooling energy by up to 40%. AI is being applied to smarter power-grid management, optimising logistics and transport, improving renewable energy forecasting, reducing industrial waste, and making buildings more energy-efficient. Software and infrastructure improvements at Google reduced energy use by a factor of 33 and carbon emissions by a factor of 44 for a typical prompt over a single year.
The overall environmental impact of AI depends heavily on how it is built, what energy sources power it, and whether the efficiency gains it delivers outweigh the electricity it consumes. In many cases, that equation is already favourable and improving.
What You Can Do as a Leader
Awareness without action is just anxiety. If you are deploying AI in your organisation, there are practical steps you can take to reduce your environmental impact and demonstrate responsible leadership.
Choose providers who prioritise renewable energy. The major cloud providers, Microsoft Azure, Google Cloud, Amazon Web Services, publish sustainability commitments and are investing heavily in renewable energy for their data centres. Where you have a choice, ask the question.
Use AI purposefully, not habitually. Encourage your teams to use AI tools where they genuinely add value, rather than as a default for every task. Complex generative tasks, particularly image and video generation, are significantly more energy-intensive than simple text queries. Thoughtful use is efficient use.
Offset your digital footprint. Many organisations already carbon-offset their travel and facilities. Extending that programme to include digital and AI infrastructure is a natural next step and a credible one.
Measure what matters. You cannot manage what you do not measure. Ask your AI and cloud vendors for energy consumption and emissions data. An increasing number provide this at the account level. Incorporate it into your sustainability reporting.
Invest in efficiency. Whether that is energy-efficient hardware, better model selection (smaller models often perform well on focused tasks), or smarter workflows, efficiency investment pays environmental and financial dividends.
Balance the ledger. If your organisation is using AI to reduce emissions elsewhere, through better logistics, smarter scheduling, reduced travel, or improved energy management — document and communicate that. The net impact of AI adoption is often positive, but only if you are intentional about it.
The Bottom Line
AI has a real environmental cost, and any leader worth their role should take it seriously. But it needs to be kept in proportion. AI currently represents around 1% of global greenhouse gas emissions, a fraction of the impact of transport, heating, heavy industry, and agriculture. The concern is not where it stands today; it is the trajectory, and whether the rapid growth of AI infrastructure is powered by clean or dirty energy.
The leaders who will navigate this well are not those who avoid AI out of environmental concern; that ship has sailed, and the competitive cost of non-adoption is real. They are the ones who deploy AI thoughtfully, demand transparency from their providers, embed AI efficiency into their sustainability strategy, and use AI itself to drive emissions reductions across their operations.
AI is not the villain in the climate story. But it is not a bystander either. How your organisation chooses to build with it, power it, and account for it will determine which side of that ledger you are on.
