AI Could Slash Emissions but Data Centres Face Pressure

AI could cut more greenhouse gas emissions than it produces — if applied responsibly across the world’s most polluting sectors. 

That’s the central finding of new research by the London School of Economics (LSE) and consultancy Systemiq, which explores how AI could transform the power, transport and food sectors – but the underlying reality is that the potential could add pressure to data centre infrastructure.

According to the study, AI technologies could help avoid up to 5.4 gigatonnes (Gt) of CO₂ equivalent emissions each year by 2035. 

Even accounting for the increased electricity consumption of data centres needed to run and train advanced models, the net impact could still be positive.

AI’s power sector potential

The power sector has long been a focal point in discussions around global emissions. Intermittent renewable energy sources like wind and solar present unique challenges that require advanced grid management, forecasting and energy balancing.

According to the study, AI-enabled tools in power generation and transmission could abate up to 1.8 GtCO₂e per year by 2035. One example already in use is at Google DeepMind, where AI helped increase wind energy’s ergonomic value by 20% by reducing the need for standby power.

The study indicates that optimising load factors and asset utilisation through AI could also increase output from existing solar and wind facilities, reducing emissions per kilowatt-hour and easing pressure on grid capacity. 

These benefits directly rely on high-performance data centres capable of real-time analysis, predictive modelling and large-scale optimisation.

AI’s demand on data centres

However, powering AI is not without environmental cost. Data centres are responsible for hosting the training and deployment of AI models, which can be energy-intensive. 

The World Economic Forum estimates that training GPT-3 alone required around 1,300 megawatt-hours of electricity.

As demand for AI services grows, so too does demand for underlying data centre infrastructure. This includes power for compute and energy-intensive cooling, particularly in AI-dense deployments where high-performance GPUs generate considerable heat.

Despite this, the report estimates that the added electricity demand from data centres running AI applications could lead to between 0.4 and 1.6 GtCO₂e in new emissions annually. This still compares favourably against potential reductions between 3.3 and 5.4 GtCO₂e per year.

The findings suggest that building more efficient and sustainable data centres is essential if AI’s benefits are to outweigh its environmental costs.

Sector-wide emissions cuts

Alongside energy, the study identifies two other key sectors for emissions reductions through AI: food and transport.

• In the meat and dairy sector, where 18% of calories result in 60% of agricultural emissions, AI can support the transition to alternative proteins. The potential abatement here is between 0.9 and 3.0 GtCO₂e per year.
  
  
• In light road transport, advances in shared mobility and electric vehicle adoption could result in 0.6 GtCO₂e in cuts, driven by AI-led optimisation of traffic, charging infrastructure and vehicle deployment.

Sophie Graham, Chief Sustainability Officer at IFS, said on LinkedIn: “From enabling smarter logistics to optimising energy grids, AI is already driving efficiency. But its impact goes further, equipping us with powerful advances in forecasting and early detection for severe weather events, critical to long-term resilience.

“IFS Industrial AI has a clear role to play in this transition, supporting asset-intensive industries adapting to a low-carbon future. Now is a pivotal moment to scale AI responsibly and equitably — especially where it can deliver the greatest climate value.”

Policy support needed

While the numbers are promising, the study acknowledges limitations. It only covers three sectors and does not assess wider system effects such as rebound emissions or economic displacement. It also excludes impacts on job creation or private investment.

“Policymakers must create enabling conditions for AI deployment, provide financial incentives for research and development, and ensure that AI applications are directed toward public goods and high-impact areas,” the paper concludes.

For the data centre sector, this means renewed focus on sustainability, cooling innovation and power sourcing as AI becomes embedded in critical national and industrial systems.