Nuclear Power Plants for AI Data Centres: a Solution to Growing Energy Challenge

Olha Zhydik

1 month ago

Nuclear Power Plants for AI Data Centres: a Solution to Growing Energy Challenge

Article

Nuclear Power Plants for AI Data Centres: a Solution to Growing Energy Challenge

The challenge of powering data centres is expanding along with the introduction of AI into everyday processes. Nuclear power plants for AI data centres are emerging as a solution to the growing need for carbon-free power sources due to the surge in AI.

The exponential growth of AI and data centres

The introduction of generative AI in 2022 sparked a technological revolution, and now, artificial intelligence is embedded in practically every sector. AI is used in manufacturing for quality control and predictive maintenance, in healthcare for drug discovery and disease diagnosis, and in financial institutions for fraud detection and risk assessment, the list of examples can go on.

Such widespread AI adoption has led to an expansion of data centres, which form the backbone of AI-driven operations and digital transformation. These facilities process and store massive datasets, which are also used for training and operating advanced AI models.

The data centers' constant power consumption raises sustainability concerns. Finding ways to achieve a balance between innovation and environmental responsibility has become more crucial as businesses continue to adopt AI technologies.

Curious about AI's evolution, the challenges it presents, and the trends shaping its future? Check out this article: Expert Insights on Generative AI: Evolution, Challenges, and Future Trends

Understanding the scale of AI’s energy consumption

The global focus on AI and data science, in general, has not only increased the number of applications of these technologies but also increased the energy consumption of the data centres that power them. Moreover, the more complex AI models become, the more energy it needs.

According to IDC, energy consumption by AI data centres is expected to reach 146.2 terawatt hours by 2027. For the context, France's total energy consumption in 2023 was 411 terawatt hours.

Key factors driving high energy use:

Processing power. AI models, especially generative AI, require extensive computational power and electricity for training. For instance, training a large language model like GPT-3 can consume around 1,287 megawatt hours, which equals the average energy needs of a household in the U.S. for 120 years.
Storage needs. The expanding volume of data from cloud services and AI will only require more storage in data centres, increasing energy use. In addition, over 30% of a data centre's energy consumption is already used for storage.
Cooling systems. Data centres generate a lot of heat, which can interfere with optimal performance, so energy-intensive cooling systems are used to prevent overheating.

Environmental implications and sustainability challenges

The environmental impact of AI and its data centres goes beyond direct energy consumption and raises concerns about the sustainability of further AI infrastructure scaling.

the amount of water the average data centre uses daily, the same amount that 1,000 American households use.

NPR

Wondering why data centres need so much water? These facilities include thousands of servers that generate heat, and water is crucial for cooling them down.

The carbon emissions picture isn't much brighter. Microsoft's latest sustainability report reveals that just building new data centres is their biggest source of Scope 3 emissions. Add to this the carbon emissions produced so that our digital world runs 24/7. While some facilities are switching to green energy, most of them still rely on fossil fuels.

The good news is that the tech industry isn't standing still, and innovative sustainability solutions are emerging to address these challenges. Let's explore them further.

How to enhance energy efficiency in AI and data centres

First, let’s talk about more efficient cooling approaches that companies can apply than traditional cooling methods.

Hot and cold aisle containment system—this strategy separates hot air produced by the data servers and air cooled by the systems. Modern energy management software can help optimise such systems by monitoring temperature levels and airflow patterns in real time.
Liquid cooling is another more efficient option to handle higher temperatures.
Using machine learning, in particular its branch reinforcement learning, to predict and regulate temperature patterns can also help to cut energy use.
Locate data centres in cooler climes. For instance, Facebook intends to place data centre in Lulea, Sweden, close to the Arctic Circle. With short, warm summers and lengthy, harsh winters, Lulea's sub-arctic environment will allow the business to benefit from natural cooling resources.

Energy efficiency can not only be achieved by reducing the amount of energy used but also by changing the energy sources. Many data centres are now trying to use solar and wind energy sources. And the hardware can be more efficient as well, for example, energy-efficient chips and servers. Developments in semiconductor technology have led to the creation of equipment with an improved performance-per-watt ratio.

Last but not least, green data centres which combine everything we mentioned above—efficient cooling, renewable energy, and optimised hardware. Moreover, some green data centres repurpose the heat generated by servers to warm nearby buildings.

Powering the AI revolution: exploring the potential of nuclear energy

US nuclear capacity has the potential to triple from ~100 GW in 2024 to ~300 GW by 2050.

US Department of Energy

Such growth being driven by artificial intelligence and data centres with a particular need for carbon-free 24/7 energy generation. Unlike unstable renewable sources, nuclear power provides consistent power generation that meets the constant needs of data centres.

The advantages of nuclear power for AI operations:

Nuclear plants provide a steady energy supply for AI operations;
A single nuclear facility can power multiple large-scale data centres;
Produces zero direct carbon emissions during operation;
Nuclear plants require significantly less land than solar or wind farms to generate equivalent power.

The production of nuclear power still faces some obstacles, including high cost of construction, strict regulations and long approval process, and community concerns, particularly from those who live close to the power station.

Industry leaders embracing nuclear power

Big tech companies are already moving toward AI operations fueled by nuclear power. Microsoft has committed to purchase 100% of the output from the revitalised Three Mile Island nuclear facility. The 837MW Pennsylvania plant is to open in 2028, once it receives regulatory approval. It is expected to power Microsoft's data centres across Pennsylvania, Chicago, Virginia, and Ohio.

Google is also planning to purchase nuclear energy from multiple small modular reactors (SMRs). This deal is expected to secure up to 500 MW of continuous carbon-free power.

In Europe, nuclear expertise is also attracting attention. EDF conducts negotiations with several major companies to power three 1GW data centres using nuclear energy. While company names remain confidential, these discussions highlight the growing interest in nuclear-powered AI infrastructure around the world.

Regulatory and policy implications for sustainable AI and data centres

The growth of AI infrastructure led to the appearance of new policies and standards, which manage energy consumption and environmental impact. The EU's Corporate Sustainability Reporting Directive (CSRD) requires large organisations to report sustainability metrics, like energy and carbon emissions. Similar reporting requirements are being introduced in the United States.

According to the Energy Efficiency Directive (EED), data centres which capacity exceeds 500 kW should report their total energy use. This report should include the amount of energy they get from renewable sources, their water usage, and how they use waste heat.

Here are a few examples. As regulations change and new ones appear, companies need to focus on sustainable practices. The success of AI will depend on balancing computing power with environmental responsibility. Because of this, following regulations should be an important part of strategic planning.

Addressing the energy demands of AI and data centres

Companies are realising the need to diversify their energy sources to power AI infrastructure. Here comes the option of using nuclear energy, as it offers steady carbon-free energy to meet the constant demands of AI operations.

While major tech companies like Microsoft and Google are already moving toward nuclear-powered data centres, the industry is also exploring other solutions, from innovative cooling systems to energy-efficient hardware.

As technology changes, regulations change too. Green data centre developments and the move toward nuclear power point to a future that may allow for further advancements in AI while upholding environmental responsibility.

Ready to transform your AI infrastructure while maintaining environmental responsibility?

Contact an expert

AI development

Partner with ELEKS to implement AI-powered strategies that drive breakthrough performance.

View service

Energy

Adopt an advanced energy management software solution to dramatically improve the productivity, accessibility, safety and sustainability of your business power system.

View industry

FAQs

Can nuclear power data centres?

Yes, nuclear power is capable of supplying energy to data centres. Modern data centres have high electricity demand to ensure uninterrupted operations. Nuclear power plants are well-suited for this because they operate continuously at high output.

Why is nuclear good for data centres?

Nuclear energy can help to boost data centre energy efficiency with uninterrupted power supply, unlike renewables, such as wind or solar sources; it is the cleanest form of power, helping to reduce the environmental footprint; it reduces reliance on fossil fuels or intermittent renewables, enhancing energy security for data centres critical to digital infrastructure.