Beware of AI! Opposition to data centres: a new risk factor 

• Alexis Bienvenu
  Fund Manager
  La Financière de l’Échiquier (LFDE) 

Northern Virginia, the world’s leading hub for data centres, has become a symbol of this resistance. Major protests have also emerged in Arizona, Texas, Pennsylvania, Michigan and New York State. Europe, India and Brazil are following suit. 

Residents are particularly critical of rising electricity prices, noise from the facilities, pressure on water resources and the transformation of landscapes into industrial zones.  

The risk is already a tangible one for investors. According to Morgan Stanley, nearly $286 billion worth of data centre projects have been delayed or cancelled between 2025 and the first quarter of 2026. The reasons are not solely down to local opposition, though this is a contributing factor, leading to construction delays and higher costs. 

At the heart of the controversies lies the issue of resources, and foremost among these is water. New data centres dedicated to AI generate enormous amounts of heat, requiring cooling systems that often rely on evaporation. This water consumption is particularly critical in a context marked by increasingly severe droughts and heatwaves. In regions already experiencing water stress, such as the western United States, opponents of data centre projects highlight the trade-off between the needs of households, agriculture, industry and natural ecosystems. 

Criticism is not limited to water. Data centres are also accused of straining electricity grids, contributing to greenhouse gas emissions and leading to land sealing. Morgan Stanley estimates that the global expansion of AI-related computing infrastructure could generate around 2.5 gigatonnes of cumulative CO₂ emissions by 2030 – a volume equivalent to around 40 per cent of the US’s current annual emissions. 

Various solutions are being sought to mitigate these negative impacts. These include, in particular, liquid cooling technologies – which use less water – water recycling, and the recovery of waste heat. Reducing the size of data centres would also help to improve their social acceptance. 

However, the fierce competition between major players – whether companies or states – leaves little time for these solutions to emerge. This is all the more true given that the construction of data centres also responds to a geopolitical urgency. Computing capacity has become a critical , just as vital to national or regional sovereignty as electricity grids or rare-earth metal mines. The rivalry between the United States and China in this area is fuelling a race for digital capacity. 

Faced with these challenges, AI investors are confronted with a choice between two paths. The first involves fully integrating the constraints relating to resources and the social acceptability of AI facilities, at the cost of what is likely to be reduced short-term profitability for hyperscalers such as Alphabet, Amazon, Microsoft or Meta, but with greater long-term resilience. The second path, which is simpler but less sustainable – and therefore more precarious in the longer term – involves shifting investment to regions that are currently less demanding, such as Malaysia, the United Arab Emirates, Saudi Arabia or Indonesia. However, there is no guarantee that this availability will continue: these territories may wish to capitalise, in one form or another, on their greater tolerance of the exploitation of their local resources. 

Taking ESG factors into account in the AI economy is therefore no longer merely a matter of social responsibility; it is becoming a key determinant of long-term value. Setting up data centres on the Moon or Mars will not change this: resources there would remain limited and protests inevitable. Resistance to data centres serves as a timely reminder that the AI economy is not merely virtual, but just as much a physical and political reality.