Answering the Boardroom Question: Can Your AI Strategy and Net-Zero Strategy Coexist?

The friction is not really about AI as a concept. It is about unmanaged growth. Running inference on an unoptimised model, defaulting to the largest available cloud instance, or duplicating data pipelines across departments all add unnecessary emissions.

Left unchecked, these habits can quietly erode progress towards science based targets.

Boards should ask executives for specifics rather than platitudes. Which workloads actually require a large model versus a smaller, fine tuned one? Are teams tracking the carbon intensity of the cloud regions they run workloads in? These are governance questions as much as technical ones.

A number of existing tools let organisations measure and reduce the footprint of their AI operations today.

Microsoft's Emissions Impact Dashboard and Google Cloud's Carbon Footprint tool both give enterprises visibility into the emissions generated by their cloud workloads, including AI training and inference jobs. This visibility is the starting point for any credible reduction plan.

Hugging Face's CodeCarbon library allows data science teams to estimate the carbon emissions of their model training runs directly within their code, making sustainability a visible metric alongside accuracy and latency.

Cloud providers have also made regional choices matter. Running workloads in data centres powered predominantly by renewables, such as Google's facilities in the Nordics or parts of its Iowa operations, can meaningfully cut the carbon intensity of the same workload compared with a grid still reliant on fossil fuels.

On the model side, smaller and more efficient architectures are increasingly viable. Techniques such as model distillation, quantisation, and retrieval augmented generation allow organisations to achieve strong results without defaulting to the largest, most energy hungry models available. Not every business problem needs a frontier scale model.

Aligning AI and net-zero strategies is ultimately a governance exercise. Leading organisations are beginning to fold AI energy consumption into their existing ESG reporting structures, rather than treating it as a separate technical silo.

This means setting internal thresholds for model size relative to use case, requiring carbon impact estimates in AI project approval processes, and holding technology teams accountable to the same sustainability metrics as facilities or logistics teams.

Some organisations have started appointing joint oversight, where chief sustainability officers and chief technology officers co-review major AI infrastructure decisions. This cross functional accountability helps prevent AI ambitions from quietly expanding scope-3 emissions without visibility at board level.

It is worth remembering that AI can also actively support sustainability goals. Machine learning models are already used to optimise energy grids, predict equipment failures before they cause waste, and improve logistics routing to cut fuel consumption.

Google DeepMind's work with Google's own data centres, for instance, used machine learning to reduce cooling energy use by notable margins. Framed well, AI becomes a lever for achieving net-zero targets rather than purely a liability against them.

Rather than treating this as a dilemma, boards should push for integrated reporting, clear ownership, and smarter technical choices.

Ask which AI investments carry genuine carbon costs, which tools are being used to measure them, and whether smaller, more efficient alternatives have been properly considered.

The organisations that get this right will not be choosing between innovation and responsibility. They will be demonstrating that the two were never truly in conflict, only in need of better coordination.