Click the logo to download your  free PDF version

           Click the logo to download your  free PDF version

 

To purchase this space contact Gordon

Data centres have become a big and fast-growing source of electricity demand. In 2024, they used about 1.5% of the world’s electricity, and International Energy Agency scenarios project that this share could reach roughly 4.4% by 2035. What does this huge demand mean for our energy system? Can it fit with efforts to fight climate change?

This study sits where two major trends meet: the rapid expansion of data centres and the phase-out of coal. These two forces pull in different directions, yet they end up in the same local electricity systems. That is where the tension becomes real—and where we most need a concrete, numbers-based answer.

Anecdotes are everywhere, but the global picture was missing

Media reports of coal plants being kept online, or having their retirement dates pushed back, to serve nearby data centres have become increasingly common. These stories are notable, but they remain individual cases. In the existing literature, most work on data centres in energy systems has focused on electricity consumption, carbon and water footprints, and the integration of renewable energy. Far less attention has been paid to how data centre demand affects the value and operation of existing fossil-fuel generation assets. What has been missing is a systematic, global assessment of whether data centre growth is broadly associated with delayed coal retirement, and if so, by how much.

What we did

To test this association, we combine 9,031 data centre facilities commissioned with 9,379 coal-fired power units worldwide. We calculate how coal units are exposed to data centres separately for four technological periods: pre-cloud (before 2006), cloud computing (2006 to 2015), large-scale machine learning (2016 to 2019), and generative AI (2020 to 2024). We use survival models to assess the relationship between data centre exposure and coal unit retirement risk, and estimate the additional CO₂ emissions associated with these lifetime extensions.

What we find

Post-2020 data centre exposure is linked to lower coal retirement risk

We find that coal units with higher exposure to data centres commissioned during 2020 to 2024 exhibit a significantly lower retirement risk, while earlier data centre cohorts show no comparable association. The pattern is driven by larger-scale facilities, including hyperscale, cloud, wholesale and cryptocurrency mining data centres, consistent with the much higher power requirements of facilities built during the generative AI era.

The relationship is concentrated among efficient coal units

We also find the association is more pronounced for units with lower CO₂ emission intensity and higher thermal efficiency. In other words, the retirement-delaying relationship is concentrated among relatively more efficient, less carbon-intensive coal units rather than the dirtiest or least efficient ones.

 Exposed data centres can emit substantial additional CO₂

Using an accelerated failure time model, we find that a one standard deviation increase in post-2020 data centre exposure is associated with about a 1.7% extension in expected coal unit lifetime. The 937 coal units exposed to post-2020 data centres can emit an additional 1.96 Gt CO₂ from 2025 until delayed retirement.

Clear implication: Net-zero pathways need to integrate digital infrastructure

The paper provides early quantitative evidence that digital infrastructure can have unintended consequences for fossil fuel phase-out. Data centres are not only large electricity consumers; their load can raise the value of dispatchable generation assets in local and regional electricity systems, particularly where grid interconnection and transmission are constrained.

At the same time, the study identifies a robust global spatial association rather than a strict causal effect, and the underlying mechanisms require further investigation. Future responses, including renewable power purchase agreements, onsite generation, storage and small modular reactors, could reshape the relationship between data centre demand and fossil generation assets.

The implication is nonetheless clear: net-zero pathways need to integrate digital infrastructure. Data centre siting, grid planning and clean-energy procurement are not separate from coal phase-out policy. If the electricity demand behind AI is left out of retirement projections, those projections may look cleaner on paper than they are in practice.