AI, Data Centres and COP31: the conversation boards need to have

There is currently an information gap around digital infrastructure and measurement of its energy consumption and overall environmental impact. No major technology company yet separates AI-specific environmental metrics from general data centre operations, and where disclosure exists it is patchy and incomplete.

While that gap persists, the technology moves on. New chips are markedly more energy efficient and are attracting serious investor money. TSMC has said energy efficiency, not raw speed, is now the top demand from AI chip customers, with power draw cut by up to 30% in newer designs while improving performance. Cooling technology is shifting too, with liquid and immersion systems now expected to cut AI data centre energy use for cooling by up to 95%. Yet most boards might not currently have or understand the difference between old and new chips, or how efficient the digital infrastructure behind their own growth plans might be.

The AI industry appears to be converging on a new yardstick – ‘intelligence per watt’ – which measures how much accurate output a system delivers for each unit of power it draws, and boards should expect it to become as familiar a figure as cost per megawatt-hour.

Boards should also be aware of the public policy debate on the social costs or ‘externalities’ of data centres. Disclosure regimes, such as the EU's CSRD and California's climate reporting law, are starting to require emissions disclosures from cloud and AI vendors, including third-party suppliers, directly into corporate accounts from 2027.

Why COP31 still matters

COP31 runs from 9 to 20 November in Antalya, Türkiye. For the first time, AI and clean technology feature explicitly in the presidency's action agenda, alongside 10 priority themes including clean energy, zero waste and methane, resilient cities and green industry. A dedicated Science, Industry and Technology Day on 16 November 2026 puts AI's dual role as both emissions source and decarbonisation tool formally on the table rather than at the fringes.

Some international progress is already being made in the effort to ensure AI is developed sustainably. The COP31 presidency announced an AI pledge on 21 September 2026, which they have stated represents a commitment on how AI is “designed, procured, powered, deployed, measured and managed” in support of climate action. On the one hand AI has huge potential to upscale renewable energy and energy efficiency, but on the other, it is also increasingly being used to drive fossil fuel production

The numbers explain the urgency. The IEA expects data centre electricity emissions to rise from around 180m tonnes of CO2 today to between 300m and 500m tonnes by 2035, depending on how fast AI-driven demand grows.

Global data centre electricity use is set to more than double by 2030 (according to the IEA). In addition, recent UN research found that everyday AI use, not model training, now drives 80-90% of that demand. The UN Secretary-General has already pressed major AI firms to disclose their full energy, water and emissions footprint, a clear sign of where pressure may be heading in Antalya.

The risk exposure for boards

The risks stack up on several fronts at once.

Strategically, AI is being adopted often for competitive reasons currently without much thought for its embedded energy, water or grid demands, and its pressures on land, nature and raw materials, leaving boards unable to answer basic questions about the cost – environmental or general – of their own productivity and growth plans.

Reputationally, community opposition to new data centre construction is intensifying from Ireland to Australia (including in the US: a July 2026 Gallup poll found 71% of Americans oppose data centre development in their communities) where local grids, water supply and consumer bills are feeling the strain.

Regulatory exposure is rising, with disclosure rules and proposed levies converging before most companies have the systems to comply credibly. There is a governance gap: boards carry growing fiduciary responsibility for AI-driven decisions, both those around ESG governance and beyond, without always having or accessing the technical grounding to challenge what they are being told, thus risking oversight that amounts to sign-off rather than real scrutiny.

The emerging response

A response is taking shape, albeit unevenly. Chipmakers and hyper-scalers are reframing the innovation race around efficiency rather than raw power and capital is following.

From 2027, California's climate law will require large companies to disclose Scope 3 emissions, which include data from their cloud and AI suppliers, pushing rigour down the supply chain whether boards ask for it or not. EU regulation is also pushing for greater transparency around how and why organisations are deploying AI, with the EU AI Act bringing in new rules for deployers of AI, from marketing chat bots to recruitment.

An emerging view among governance advisers is increasingly urging that AI oversight and ESG oversight sit within the same committee/board so the board overseeing the AI strategy is the same board questioning its footprint; not two conversations running past each other.

Despite these efforts, there has not yet been a commensurate increase in environmental impact transparency.

Finally, and in line with a recent RAND paper on strategy during deep uncertainty, perhaps the wisest board posture is to preserve optionality by building the technological skill, dataset and governance structures now to respond credibly whichever way the technology, regulation and environmental situation evolves.

Key questions boards could ask

Good oversight does not require every board director to become an AI or energy specialist. It requires knowing which questions to ask and having the standing to expect a considered answer.

• Can management show us, in practice rather than in principle, the energy, water and emissions footprint of the AI and digital infrastructure we depend on, including what we buy in from third-party cloud and AI vendors?
• Are we ready for the emerging data centre levies and disclosure regimes taking shape, such as CSRD?
• Do we have a clear line of sight to assess additional watts of AI-driven energy and its relation to cost or the efficiency it delivers?
• Whose job is it, on this board or in management, to own the environmental consequences of our AI and digital strategy, and is that scrutiny holding up to the same standard we apply to financial and climate risk?

Do we have a clear accountability chain for how AI is being deployed in our own disclosure, reporting, and decision-making?

This article follows and updates Sustainability and AI: what boards must decide now. The views expressed in this article are those of the authors alone and do not necessarily reflect the views or positions of their affiliated organisations.