From Threat to Pillar: Reframing AI's Carbon Footprint as a Sustainability Strategy
I see the publishing industry making huge strides in its physical sustainability footprint. In my own organisation, Hachette UK, for example, I’m proud to see our climate targets officially validated by the Science Based Targets initiative (SBTi), and our distribution centres achieving zero waste to landfill. This is the kind of deep, measurable commitment that is rightly celebrated.
As we work diligently to reduce our physical impact, we are simultaneously embracing technologies like Large Language Models (LLMs) and AI. For years, this created a worrying paradox: our physical sustainability was improving while our digital sustainability was, by all accounts, set to degrade, with reports of AI’s significant energy and water demands.
So, how do we reconcile our green mission with our digital ambition?
My view has always been that we should not see AI as a threat to our sustainability goals. We must reframe it as a core strategic pillar of our total sustainability plan. Recent data has only strengthened my conviction. The narrative of AI as an unavoidable energy-guzzler is being replaced by a more nuanced, manageable reality. This requires us to manage two distinct footprints: the physical and the digital.
Demystifying the Digital Footprint
For a long time, the debate was dominated by abstract fears and estimates that a single generative AI query used perhaps ten times the energy of a standard Google search. That figure understandably caused alarm.
However, recent, transparent reporting from the hyperscalers themselves paints a dramatically different picture. Google’s August 2025 report on AI energy consumption provided a crucial benchmark: a median Google Gemini query consumes just 0.24 watt-hours (Wh) of electricity.
To put that in perspective, it’s roughly equivalent to running a microwave for one second. This figure is not only not 10x a standard search (often estimated around 0.3 Wh), but in the same ballpark, and vastly more efficient than early models. This is thanks to incredible, rapid gains in efficiency - Google reported a 33-fold improvement in its model efficiency in the 12 months to May 2025.
As a comparison, 0.24 Wh, is around 0.03g of CO2. Instagram generates around 1g CO2 per minute of scrolling and TikTok: 2.63g! Checkout this CO2 calculator based on Greenspector’s analysis.
This isn’t a case of accepting a necessary evil. It’s an opportunity for smarter, more holistic environmental stewardship. Here’s a pragmatic framework for how I believe we can approach this.
The Pragmatic Approach: Using AI Intelligently
Before we can manage AI’s impact, as leaders we must be clear-headed about what we are using and why.
1. Use the Right Tool for the Right Job. The new data makes this point non-negotiable. Not all “AI” is equal. While a simple text query is now remarkably efficient, recent research highlights that generating a single image can consume over 10x that energy (around 2.9 Wh), and generating a short video can require as much as 90 Wh.
The first step for any leader is to demystify the technology. We must ensure our teams aren’t using the sledgehammer of a full-blown video generation model when a simple text query (at 0.047 Wh) or a traditional machine-learning model will do.
2. Use AI for Greater Sustainability. This is the most powerful part of the reframing. We can and must deploy AI to drastically reduce our physical waste. I’m excited by the potential of using AI to extract meaningful data for predictive analytics, fine-tuning print runs, and minimising the production of ineffective marketing materials. We can use it to further optimise logistics from distribution centres, reducing fuel consumption.
The tiny carbon cost of a 0.24 Wh query is dwarfed by the carbon savings of preventing a single lorry from making an unnecessary journey.
3. Demand Smart Infrastructure and Green Partnerships. The low energy figures we’re seeing aren’t magic. They are the result of colossal investment in custom chips (like Google’s TPUs) and renewable energy. We are not building these models from scratch. We rely on infrastructure partners, and we must hold them to account. When selecting cloud providers, it’s crucial to scrutinise their commitments to renewable energy and their (Google’s) 2025 Environmental Reports (Microsoft’s is also available, and Amazon’s is here). Our digital supply chain requires the same rigorous environmental standards as our physical one.
Governance and Measurement: Managing the Total Impact
There is a timeless management principle I love to use: What is measured, is managed. This connects to a broader challenge in AI strategy — balancing quick wins with long-term transformation, where sustainability measurement becomes part of that strategic balance.
To treat our digital footprint as a strategic concern, we need a governance framework.
1. Calculate the Net Impact. The term “AI” is broad, and so a crucial task is to categorise our AI usage. From there, we must establish metrics that calculate the net environmental effect. The equation is simple in concept and, thanks to new data, becoming more achievable in execution:
Net Impact = (Carbon Savings from AI-driven Outcomes) - (Carbon Cost of AI Process)
Calculating this precisely is still an emerging discipline, and yet a pragmatic first step is to leverage the carbon footprint dashboards now offered by most major cloud providers (AWS, Azure, Google Cloud). This allows leaders to pilot the measurement process, creating a tangible business case.
2. Look Beyond Carbon: The Water Footprint. While the carbon cost per query is proving manageable, the water footprint is emerging as the next critical challenge. Data centres consume vast amounts of fresh water for cooling. Research has estimated that global AI demand could consume (a paper from researchers at the University of California, Riverside) a staggering amount of water by 2027 - perhaps as much as half the UK’s total annual water withdrawal. Other studies suggest a handful of complex queries can consume a 500ml bottle of water.
This is where governance becomes vital. Our digital strategy must account for water usage effectiveness (WUE) as well as carbon, especially when deploying models in water-stressed regions.
3. Create a Cross-Functional Dialogue. This isn’t just a matter for the tech team or the sustainability officer. True governance requires a conversation between IT, Finance, Operations, and Sustainability leaders. Who should own the ‘carbon and water budget’ for AI? How is it balanced against departmental P&Ls? Establishing this cross-functional forum is essential for turning strategy into practice and for developing guidelines that empower all staff to make sustainable choices.
By adopting this strategic mindset, we can transform AI from a shadowy threat into a powerful and measurable pillar of our sustainability mission.
What is the single biggest hurdle your organisation faces in balancing its digital ambition with its green commitments?