What is the environmental impact of AI? 

AI isn’t an abstract cloud floating somewhere; it’s deeply physical. Behind every prompt are data centres full of servers that need to be powered, cooled, repaired, replaced, and eventually discarded. Some of the key areas that matter when we talk about AI’s environmental impact include: 

1. Energy: LLMs require a lot of electricity. By 2030, AI could account for 4–6% of global electricity demand – a huge share for a single technology. And asking ChatGPT something uses 10–20 times more energy than running a standard Google search.
2. Hardware: Training and running models requires a steady churn of high-spec chips and machines. Between 50–80% of a data centre’s lifetime emissions come from this hardware, and AI is expected to contribute to around 12% of global e-waste by 2030. 

In other words: AI has a significant environmental footprint, and it’s growing. 

So what can we do about it individually? 

Our team explored a set of simple principles for environmentally responsible, lower-impact AI use that anyone can adopt: 

1. Use the smallest model that gets the job done: Not everything requires a heavyweight model. For quick summaries, writing tweaks, or simple analysis, lighter models like Gemini Flash, GPT-4o Mini, or Claude Haiku do the job beautifully – and with a fraction of the energy use. Save large models for genuinely deep, nuance-heavy tasks.  
2. Prompt intelligently: Clear prompts = fewer tokens = lower environmental cost. Adding constraints (“keep to 150 words”, “give three bullet points”) or batching questions into one message reduces unnecessary processing.  
3. Avoid wasteful habits: We all know the temptation to hit “regenerate” because we’re curious. But this is the AI equivalent of leaving the tap running. Keeping conversations in a single thread, pasting only what’s needed, and avoiding multiple drafts unless necessary all help cut down impact.  

What can governments and system-level leaders do? 

While individual users and organisations can make meaningful changes, the biggest opportunities sit at a national or system level. Public-sector leaders, regulators, and policymakers can shape the sustainability of AI far upstream of day-to-day usage, by taking four actions. 

1. Plan for sustainable energy generation for AI 

AI growth zones, national compute clusters, and public-sector AI platforms will require huge amounts of electricity. Governments can aim to: 

• Align AI expansion with clean energy strategy, ensuring compute demand is met by renewables, grid upgrades, or emerging options such as small modular nuclear reactors (e.g., proposals around Wales’ AI Growth Zone). 
• Incentivise energy-efficient compute through procurement standards and tax incentives. 
• Require transparency from cloud providers on carbon intensity, allowing public-sector buyers to choose lower-impact options. 

2. Manage where and how data centres are built 

Placement of data centres has real environmental consequences. Governments can: 

• Assess local water systems to avoid placing high-water-use facilities in water-stressed regions. 
• Protect biodiversity and local ecosystems by embedding habitat impact assessments in data-centre planning approvals. 
• Set minimum performance standards for cooling technologies, requiring more efficient or water-free approaches where viable. 
• Use spatial planning to cluster compute in areas with sustainable energy access, grid capacity, and low ecological sensitivity. 

3. Reduce the embodied carbon of AI hardware 

Model training depends on constant hardware refresh cycles, which carry significant environmental costs. Policymakers can: 

• Mandate circular-economy practices for chips and servers: repair, reuse, refurbishment, and responsible recycling. 
• Support UK-based recycling and materials recovery, reducing dependency on overseas processing with high environmental impacts. 
• Require lifecycle carbon reporting from cloud providers as part of procurement. 

4. Build sustainability into national AI governance 

Finally, sustainability should be a core part of AI governance, not a bolt-on. This might include: 

• Incorporating environmental metrics into national AI assurance frameworks. 
• Including environmental criteria in public-sector AI procurement exercises. 
• Supporting research into low-resource models and energy-efficient architectures. 

Together, these whole-system levers can dramatically reduce the environmental footprint of AI adoption across the UK. 

Why this matters for public sector leaders  

Many of our clients across the UK public sector are exploring how to adopt AI in a way that’s effective, safe, and aligned with their values. Environmental sustainability is increasingly part of that conversation. There’s growing recognition that AI’s environmental footprint isn’t something to consider as an add-on: it should shape digital strategy and procurement choices from the outset. 

What we’ve learned internally is that the most meaningful shifts often start with awareness and skills, not grand strategies. Once people understand the impact of model choice, prompting behaviour, and usage patterns, they naturally make more thoughtful decisions.  

That’s one reason we’re thinking about how we can support public sector teams in building exactly this kind of capability, with practical upskilling that helps people use AI well, not just more.  

If you would like to discuss more about sustainable use of AI, please contact josh.myers@thepsc.co.uk.