Artificial intelligence drives progress, but its energy demands come at a cost. A single ChatGPT query uses about ten times the energy of a typical Google search, which consumes around 0.3 watt-hours. Training GPT-3 consumed 1,287 MWh and consequently emitted 502 tons of CO₂ – similar to driving a car to the moon and back.
AI’s reliance on electricity could reach unprecedented levels. By 2030, AI is expected to account for around 5% of Europe’s total electricity use, up from the current 2%. As AI grows resource-intensive and emissions increase, companies must turn to carbon-free energy sources.
Small modular reactors present a promising pathway to sustainably meet AI’s energy demands while addressing regulatory and scalability challenges. Despite their potential, SMRs are still in the early stages of development, currently at technology readiness levels (TRL) 5–6. The deployment of the first small modular reactors is expected by 2030 at the earliest.
What are small modular reactors?
A small modular reactor (SMR) is a nuclear reactor that produces energy through nuclear fission. In this process, the nucleus of a heavy atom, like uranium, splits into two smaller parts, releasing a large amount of heat. The produced heat is then used to generate electricity.
SMRs are smaller and simpler than traditional nuclear plants. They are made from compact, factory-built parts that can be assembled as needed. SMRs can power cities, provide heat for factories, or turn seawater into fresh drinking water.
Why are small modular reactors important?
SMR’s parts are factory-made, faster, and cheaper to build, reducing labor and construction time to as little as three years, compared to up to 12 years for traditional plants.
Compact SMRs, for instance, can be used in different locations (e.g., small markets, isolated areas, and places with limited water). In addition, SMRs can be scaled to match energy needs, replace old plants, and work alongside other zero-emission sources.
SMR projects can boost nuclear energy, potentially creating 7,000 jobs and generating over $1 billion in sales for a 100-megawatt plant.
Why does AI consume so much energy?
AI’s energy demands start with the massive energy needed to train large models. GPT-4 requires processing vast datasets through extensive calculations across thousands of high-performance servers. Training GPT-3 used around 1.3 GWh of electricity, equivalent to fully charging over 100 million smartphones.
Data centers engage in intense data processing, which is supported by:
- Servers performing complex computations
- Storage systems managing large amounts of data
- Cooling systems ensuring operational stability by preventing equipment from overheating
All of this requires substantial energy. It is estimated that data centers use ~1-1.5% of the world’s electricity, with data center energy consumption expected to rise as AI grows.
AI applications, such as ChatGPT, must provide instant responses to users, meaning they must continuously consume power, even when idle. As AI models become more complex, their energy needs increase quickly. The computing power for AI doubles about every 100 days, showing just how fast these demands are escalating.
How can small modular reactors meet AI’s increasing energy demands?
SMRs provide a stable 24/7 power source, ideal for the continuous demands of AI data centers. SMRs have capacities ranging from 20 to 300 megawatts, meeting energy needs without large infrastructure, as in the case of traditional nuclear plants.
Independent of power grids
Small modular nuclear reactors operate independently of centralized power grids, ensuring resilience for AI-critical functions. This self-sufficiency reduces risks from outages or grid instability, allowing continuous operations. Such features are essential for the uninterrupted processing capabilities of AI technologies.
Low carbon footprint
SMRs present a low-carbon alternative to fossil fuels, aiding tech companies in achieving sustainability objectives. For example, in Canada, SMRs could reduce emissions by 19-59 Mt by 2050, a 3%-9% decrease from 2020, mainly in oil, gas, and manufacturing industries.
Faster deployment
The modular nature of SMRs allows for quicker construction times (3-5 years) compared to traditional nuclear reactors (6-12 years). Also, factories produce modules that are assembled on-site, reducing construction delays and costs. This rapid deployment capability becomes a key aspect as demand for energy rises sharply.
Ideal for remote locations
Their compact design allows SMRs to power remote AI facilities without traditional grid infrastructure. This capability supports technology companies expanding into less-developed regions with minimal energy systems.
Reliable safety features
Modern small modular reactor designs include advanced safety features, lowering accident risks and boosting public confidence in nuclear power. SMRs use passive cooling, low power, and low pressure, reducing accident risks. Many need less refueling; some can operate up to 30 years without it.
Key challenges for scaling AI:
The International Atomic Energy Agency has identified over 70 SMRs designs worldwide, each offering different safety and efficiency features. To make SMR production affordable, standardizing a single design is essential to reduce production costs.
The utilization of SMRs faces a regulatory mismatch across countries. Current regulations were designed for traditional nuclear reactors, which need to address SMRs’ unique features, which will slow deployment. There is a need to harmonize licensing requirements across countries.
A robust supply chain is needed since SMRs’ new technologies and manufacturing processes still need to be fully developed, risking delays and cost overruns. Additionally, supply chain inflation increases costs and complicates timely delivery.
Opposition to nuclear power plants for AI data centers means engaging communities early, addressing safety concerns, and involving local stakeholders to build trust and acceptance.
The first-of-a-kind builds can be costly, and without clear demand signals or community support, private capital remains hesitant to invest in new nuclear projects. Addressing these financial uncertainties through policy recommendations could enhance the bankability of SMRs.
Global adoption of small modular reactors:
Here, we examine the efforts by various organizations and governments to adopt small modular reactors (SMRs).
Google signed a corporate agreement with Kairos Power to use multiple SMRs for 24/7 carbon-free energy. The goal is to power Google’s data centers and offices with clean, reliable nuclear energy.
The first SMR is set to launch by 2030, with more added through 2035, supplying up to 500 MW to the U.S. grid. Kairos Power’s reactors use a molten-salt cooling system and ceramic, pebble-like fuel, creating a low-pressure nuclear solution.
Amazon
Amazon entered into a series of agreements to support nuclear energy through SMRs. The aim is to power Amazon’s operations and regional grids sustainably.
In Washington, Amazon is partnering with Energy Northwest to develop four SMRs, starting with 320 MW and potentially expanding to 960 MW. Amazon is also investing in X-energy, which will supply advanced SMR designs and manufacturing for up to 5 gigawatts of nuclear energy. Lastly, in Virginia, Amazon’s agreement with Dominion Energy aims to add 300 MW from SMRs to support local energy needs.
Microsoft
Microsoft has signed a 20-year Power Purchase Agreement (PPA) to procure 835 MW from a revamped Three Mile Island nuclear power plant. The objective is to support the energy demands of its AI-focused data centers. This agreement reflects Microsoft’s strategy to secure a reliable, carbon-free energy source, particularly as AI workloads increase energy consumption.
US Government
The US Department of Energy (DOE) has launched a $900 million funding program to support the development of SMRs and other advanced nuclear technologies. The funding aims to accelerate SMR deployment as part of the nation’s clean energy strategy, promoting low-carbon, reliable power sources. The program is open to companies and public utilities that can help replace coal plants and support regional energy needs.
Turkey
Turkey is drafting legislation to support the construction of SMRs as part of its push towards a diversified, sustainable energy mix. This new law aims to streamline the regulatory framework for SMRs, making it easier for domestic and international companies to develop these advanced nuclear reactors.
TerraPower, BWXT Technologies, NuScale
BWX Technologies (BWXT) has been awarded an engineering contract from TerraPower to support the Natrium™ Demonstration Project. This project is an advanced reactor initiative with significant capacity and flexibility.
The Natrium design features a 345 MW sodium-cooled fast reactor with molten salt energy storage. It can increase output to 500 MW during peak demand, sufficient to power approximately 400,000 homes.
The project is supported by a $1.9 billion commitment from the US DOE under the Advanced Reactor Demonstration Program. The project aims to operationalize the reactor by 2030.
Nuclear vs renewable energy: What is the most viable pathway?
France and Germany illustrate two different paths for meeting the energy demands of AI-powered data centers. France is investing €1 billion in small modular reactors by 2030, positioning nuclear as a stable, weather-independent source ideal for AI’s constant power needs. This promises reliable energy but faces challenges with costs and nuclear waste.
Due to the downsides of nuclear power, Germany officially ended its use of nuclear power with the shutdown of the last three reactors on April 15, 2023. Germany now focuses heavily on renewables like wind and solar. While environmentally friendly, renewables depend on the weather, creating reliability issues for the 24/7 operation of AI data centers. Addressing these gaps would require costly updates to Germany’s energy infrastructure.
Choosing nuclear and renewables involves balancing reliability, scalability, and environmental impact. A hybrid approach – nuclear for consistent power and renewables for peak support may offer the most effective solution, combining stability with sustainability.
Will AI continue to boom with nuclear energy?
The rapid growth of AI is driving increasing energy demand, particularly in data centers. AI-focused facilities require immense computational power for model training, often scaling to gigawatts of energy. Such demand strains existing grid infrastructure, requiring an urgent need for alternative energy solutions.
Renewable resources such as wind and solar remain vital for reducing carbon emissions, but their intermittent nature makes them insufficient for tech industries requiring continuous, reliable power. To address this major and immediate concern, companies are turning to “clean capacity” – energy sources that are both sustainable and reliable.
Thus, nuclear energy is emerging as the ‘key solution’ to support the scaling of AI. It can provide firm, consistent power necessary for achieving carbon-free energy goals by 2030. However, its long-term viability as a sustainable energy source depends on addressing key challenges around nuclear waste management, safety, cost overruns, regulatory hurdles, and public perception.
