Is AI bad for the environment?

Summary: Individual AI use is not where the environmental story lives. This post draws on Spark AI's research to show where the impact actually sits, and what intentional adoption looks like in practice.

Date: 18.03.26

Read Time: 4 minutes

Spark AI is a strategy-led consultancy helping agency and brand teams move from fragmented experimentation to organization-wide capability. Our blog provides the strategic techniques, insights and industry discussions needed to navigate AI with confidence. 

Should agencies feel guilty about using AI?

One of the most common questions we get asked is about how sustainable AI actually is. The anxiety is understandable given the headlines.

Last year, Spark AI co-founder Emma Wharton published a detailed analysis of AI's environmental impact. What she found might surprise you. 8 prompts a day for an entire year uses the same energy as running a space heater for 2 hours. 50 queries a day, for the rest of your life, emits less CO2 than a 1 hour-long commercial flight. Individual AI usage simply does not carry the footprint most people assume it does.

So where does the real impact sit?

Consumer chatbots account for just 3% of AI's total energy consumption. The remaining 97% is embedded in platforms most of us use without a second thought: the sponsored post on Instagram, the contactless payment screened for fraud, the Netflix thumbnail chosen for you. These systems have been running continuously, at enormous scale, long before generative AI entered the conversation.

That matters. Because it changes where agencies should actually focus their attention.

Where does the sustainability opportunity lie?

When AI is embedded into production workflows, it can actively replace some of the most carbon-intensive parts of the process. Virtual prototyping and concept visualisation reduce the need for location scouting and crew travel. Fewer physical reshoots means a tangible reduction in an agency's overall footprint.

The more useful question is not "how do we limit our AI use?" It's "where can AI remove the most carbon-intensive work from our process?"

How should teams be using AI day-to-day?

How teams prompt matters more than most agencies realise. Our ‘Think, Prompt, Think’ framework was built to break the cycle of vague, repetitive prompting that produces poor results and wastes compute energy.

It works in three steps. Think first: define the strategy or brief before opening any tool. Prompt deliberately: craft requests with specific context and clear constraints. Think again: apply human judgement before you iterate. Knowing exactly what you are testing means you make fewer requests and reach better outcomes faster.

This is where AI literacy makes a real difference. A team that does not know how to use AI well will use more energy reaching outcomes than a skilled team could achieve in far fewer attempts. Capability is not just a performance issue. It is an environmental one.

Does it matter which AI vendors we choose?

Yes. AI infrastructure does not carry the same environmental cost across providers. Ask your vendors for carbon metrics. Find out where their data centres are located, what energy sources they run on, and whether they offer green compute options. These are reasonable questions, and any vendor worth working with should be able to answer them.

Prioritising providers who publish meaningful environmental data and invest in renewable infrastructure is where agency-level decisions start to move the needle.

Will the data stay the same?

While the findings hold true now, the environmental picture around AI is shifting quickly. Figures on energy consumption, carbon emissions, and infrastructure investment are changing as the technology scales and as more research is published. The data in this post reflects what was known at time of writing. We recommend treating specific numbers as directional, and checking primary sources when making decisions.

FAQs

Will using AI eventually make our agency more sustainable than traditional methods?

It can, if AI is used to genuinely reshape the operating model. Replacing carbon-intensive activities like global location scouting with AI-driven visualisation and high-fidelity prototyping can substantially reduce the total environmental cost of a project. That outcome takes deliberate adoption, not simply access to tools.

Is the energy cost of training a custom AI model worth it environmentally?

For many agencies, yes. A custom-trained model is often more efficient at specific tasks than a general-purpose one, which means fewer prompts and more accurate outputs first time. The energy saved from fewer failed attempts frequently outweighs the upfront training cost.

How do we start measuring the environmental impact of our AI usage?

Start with your vendors. Request carbon metrics and ask which green compute options they offer. Alongside that, track the reduction in physical travel and production waste that AI adoption enables. Those figures show net-positive impact most clearly.

Turn fragmented AI experimentation into organisation-wide AI capability – with impact, control and confidence.https://www.wearespark.ai/