Small modular reactors alleviate data centre power demand

How small modular reactors can help alleviate increasing power demand from data centres

Data centre power consumption is becoming a key climate policy challenge. The surge in demand is being driven by the widespread use of artificial intelligence models, machine language models, large language models and cloud services. There are indications that the increased demand is being met by natural gas and, in some cases, diesel is used to back up the server base loads. Available predictions of power demand surge indicate there is an urgent requirement to address the issue. One of the practical solutions is the deployment of small modular reactors. Utilising small modular reactors offers a more sustainable alternative energy source.

A key step in artificial intelligence machine language model development is model training. In artificial intelligence and machine language projects, model training involves processing large datasets, equations and scientific interpretations to a specific end use.

For example, a pharmaceutical company testing for a new drug formulation. This step is time-consuming and power-intensive and currently generates substantial CO₂ and non-CO₂ emissions. Model training takes hours, and in some cases weeks, generating emissions that are roughly equivalent to the emissions of a trans-Atlantic flight [1]. This increased demand for power is being met by power grids that are fed by the combustion of natural gas. It is vital to understand that this additional power demand will impact the available carbon budget. The Sixth Assessment Report [2] projected that the remaining carbon budget for the 1.5⁰C target is 500 GtCO2. At present, the world's CO₂ emissions are greater than 30 GtCO₂ per year [3].

The deployment of small modular reactors (SMRs) to power up the data centres offers a sustainable solution to meet additional demand and offers the following advantages:

• Generate sufficient energy to meet the data centre power requirements.
• Meets a key decarbonisation criterion of additionality. The low carbon generation from small modular reactors will be both additional and permanent for the duration of the project.
• Increase the likelihood of meeting the requirements of the carbon budgets in the year 2100.
• Avoid dependence on using temporary solutions that range from buying carbon credits to emissions reduction attempts through geo-engineering.

TÜV SÜD is able to support SMR development and data centre industries through the following services:

• Energy modelling and long-term demand balancing
• Computational fluid dynamics modelling for cooling
• Energy balancing in manufacture of decarbonised HALEU fuel for small modular reactors
• Environmental auditing with ISO 50001 support
• Workshop management to develop knowledge value chains
• Gate papers for sustainability optioneering
• Developing metrics for power usage effectiveness
• Research and development

Future Outlook

To attract investment and gain global share in the data centre market, the energy generation policy in geographical hotspots is already changing. For example, in Sweden, a Data Centre campus powered [5] by a small modular reactor was proposed in 2023. In the United States, a small modular reactor data centre campus is being discussed in Virginia [6].

Within the UK, small modular reactors have some momentum and investment drivers, however, direct intervention from the government is required to engage in a specific data centre campus programme supported by SMRs to meet the surge in power demand. Evidence from data centres in the United States clearly indicates that data centre power usage effectiveness significantly declines over a period of time [7]. In terms of energy generation policy, this means a secure low-carbon power source is needed to power the data centres.

Find out more about our services for new nuclear technologies.

For our latest news and updates, subscribe to our newsletter and follow us on LinkedIn.

Written and reviewed by:

Arshad Bhat, PhD and Megan Davies, MSc

References

1. Strubel et al. Available at [1906.02243v1] Energy and Policy Considerations for Deep Learning in NLP (arxiv.org)
2. Sixth Assessment Report. Available at Sixth Assessment Report — IPCC
3. Global Carbon Budget. Available at ESSD - Global Carbon Budget 2023 (copernicus.org)
4. Electricity Analysis and Forecast 2026. International Energy Agency report. Available at Electricity 2024 - Analysis and forecast to 2026 (iea.blob.core.windows.net)
5. Karnfull Next. Available at SMR Campus | Kärnfull Next (knxt.se)
6. Green Energy Partners. Available at Green Energy Partners partners with nuclear power provider IP3 for Virginia project - DCD (datacenterdynamics.com)
7. Powering Intelligence. Analysing Artificial Intelligence and Data Center Energy Consumption. Available at EPRI Home