Climate risk and data centres

By Rebecca Delaney

Aug 14 - (The Insurer) - As the data centre boom drives stress on local water supplies and ecosystems, facilities are now themselves increasingly vulnerable to the impacts of climate change-related risks as escalating weather events raise costly business interruption (BI) exposures.

Accelerated demand for data centres globally is being driven by AI, cloud technology and digital sovereignty needs.

In January, British Prime Minister Keir Starmer’s AI opportunities action plan set out steps to accelerate the build-out of data centres around the UK, while commentary by Moody’s Ratings last month projected that data centre capacity across Asia Pacific will more than double through 2030 to account for over 40% of global capacity.

However, academics and others have voiced concerns around the environmental liability risks posed by data centres.

A report on July 21 by the Independent Water Commission, chaired by former deputy governor of the Bank of England Jon Cunliffe, recommended the abolition of UK water regulator Ofwat in favour of a new body to more appropriately monitor the industry, with rising data centres identified as one of the “pressures” putting “huge demands” on the UK’s water system.

A significant amount of water is needed to cool hardware used for training and deploying generative AI models, which can strain local water supplies and disrupt ecosystems.

In addition, research by Morgan Stanley in September last year projected that the global data centre industry will emit 2.5 billion tons of CO2 through 2030, as a result of the computational power required to train GenAI models.

“It's important to highlight that data centres pose a risk to the environment in terms of impact to water scarcity, pollution, soil contamination, and water contamination,” Ester Calavia-Garsaball, senior director, physical risk at WTW’s climate practice, told Sustainable Insurer.

“This is quite a new area which is less understood and quantified, but we are working with Willis Research Network on a current workstream project looking at the environmental risk posed by data centres and how they can be better understood, quantified and managed.”

The project explores the full lifecycle of a data centre in the context of AI proliferation, from design and construction to operation and decommissioning.

It will assess environmental externalities (such as water usage, carbon emissions and air pollution) and identify emerging liability risks linked to data centre placement, energy sourcing and waste disposal.

In an ironic self-perpetuating cycle, data centres are now increasingly vulnerable to weather events exacerbated by climate change, with downtime looming as an expensive possibility.

Data centres are located globally in regional concentrations (for example, northern Virginia has emerged as a hub in the U.S.) based on factors such as land availability, proximity to power infrastructure and transportation, and local regulations.

“By connecting multiple data centres into a cluster, they can act as a backup for each other. If one data centre experiences an outage, the others can ensure continuous operations to minimise downtime,” Calavia-Garsaball explained.

“Large catastrophic events, like an earthquake or hurricane, are a particular threat because they can impact multiple locations at the same time. If a catastrophic event impacts various locations of the cluster, then they have a significant issue and can have physical damage and downtime.”

Although property damage remains a risk to data centres, the associated BI from any downtime is likely to have a greater financial impact, she added.

As well as large cat events, clusters are vulnerable to more localised weather events, such as flooding and windstorm, which could also impact the infrastructure and utilities that data centres rely on, such as electrical substations and hyperscalers.

“Chronic long-term risks – like sea level rise, drought, heat stress, and temperature changes – are also important for data centres because they will impact things like availability of water for the cooling system,” Calavia-Garsaball continued.

“Heatwaves and temperature thresholds will impact if and when data centres need to start activating the cooling system, which will impact operational expenditure, or even whether they can operate at all.”

Accurately modelling the financial impact of downtime of technological infrastructure is difficult, an issue that has been encountered in the cyber insurance market.

For data centres, modelling downtime resulting from climate risks is based on the expected intensity of a given peril or hazard, for example, anticipated flood depth or earthquake strength.

“Downtime estimates can be calibrated with actual experience of looking back at real events. It's always good to calibrate the more deterministic downtime estimates with real-life events and actual losses and claims,” said Calavia-Garsaball.

“Climate mitigation and adaptation measures play an important factor in quantifying downtime, as well as business continuity planning and business preparedness agreements with power and infrastructure backups.”

With demand for data centre capacity showing no signs of slowing, the insurance industry’s ability to quantify and model BI exposures remains crucial as physical climate risk increasingly permeates digital boundaries.

“We're seeing at the moment that BI is often what drives losses for nat cat or climate, so we're paying more attention to how it's being understood and quantified in the models,” Calavia-Garsaball concluded.

“It's often underestimated in some of the models, so that's a growing area for us to make sure we address this. It's pretty industry driven – in the 10 years I've been in the team, I see more and more focus on innovation on how we understand BI.”