Since the rise to prominence of ChatGPT just two and a half years ago, the adoption of AI tools by businesses, governments and individuals globally has grown exponentially. Recent research suggests the global LLM market was worth nearly $6.33bn in 2024 and is expected to grow to $25.22bn by 2029, a compound annual growth rate of 31.83%. Survey data from Elon University meanwhile suggests more than half of Americans now regularly use LLMs, a remarkable statistic given the relative novelty of this technology.
The enthusiasm for this technology however is tempered by significant questions about how to power it. Recent research published in Nature estimated that “implementing ChatGPT-like AI into every Google search would require between 400,000 and 500,000 NVIDIA A100 servers”. This would require 23–29 TWh of energy annually – 23–30 times the energy of a normal search.
When it comes to delivering power at this scale, and sustainability, nuclear energy has quickly become the favoured solution for powering data centres and the AI revolution. It’s here where things get complicated, however. There are a range of different models available to players in the industry, as well as a range of financial, regulatory and delivery challenges to be met.
Chain reaction
Nuclear energy dominated headlines and data-centre hyperscalers’ press releases in the back-end of 2024, with a number of models being trialled to help AI go nuclear.
Co-location with existing large reactors (i.e. the Talen-Amazon Web Services data centre deal) and Power Purchasing Agreements (PPA) underwriting the restart of dormant reactors (i.e. Constellation Energy’s revival of Three Mile Island for Microsoft) both highlighted approaches available for the current light-water reactor fleet. Alongside this, though, hyperscalers have shown themselves to be open to advanced reactor technologies as well.
Recent notable corporate agreements have included Google and Kairos Power’s partnership on the delivery of up to 500 MW of power from up to seven molten salt reactors, with expectations that the first reactor would be running by 2030 and the full fleet complete by 2035. Google has also entered into an agreement with start-up Elementl Power to develop three sites for advanced reactors, with each site expected to generate at least 600 MW. Upon project completion, Google will retain the option to purchase power from the reactors.
Amazon, meanwhile, recently announced a co-investment in X-energy’s US$500m Series C funding round, further committing to an initial 320 MW Small Modular Reactor (SMR) project with Energy Northwest in Washington State using X-energy’s design. The project includes an option to increase the installed capacity to 960 MW.
Not to be forgotten, Meta has announced a request for proposals (RFP) for 1-4 GW of new nuclear capacity, with eligibility extended to both traditional large reactors as well as SMR projects.
The time is right?
While most power purchase agreements signed by hyperscalers for supply by advanced reactors acknowledge start dates in the 2030s, questions nevertheless remain as to whether current players can deliver working reactors as per this demanding schedule.
Our own conversations with experts from the industry suggests that deploying next-gen reactors within the next decade will be challenging, and such reactors are unlikely to be cost effective unless less proven reactor technologies (such as molten salt reactors) are used to limit upfront CAPEX.
Furthermore, the actual revenue rates on power needed to underwrite an advanced reactor for data centres are likely to be significantly above current industrial energy pricing. Some of the bottlenecks identified by industry experts including permitting, affordability and supply chain constraints. One of our experts estimates a realistic timeline for mass deployment of SMR technologies to be 15-20 years, with prototypes taking 7-10 years to provide proof of concept.
Other experts have estimated, with a high degree of confidence, that 20 SMRs within 20 years is a feasible timeline for what type of nuclear capacity can be added based on existing supply chains. On the more optimistic side of the spectrum, some experts anticipate the earliest commercially active SMR in the US to be operating by late 2030 / early 2031, dependent on regulatory approvals.
Counting the cost
Another key factor likely to determine the shape of the industry emerging around data centre deployment will be price. New SMR generation could feasibly cost $7,000-8,000/kW installed, with any reactor designs that boast $2,000-3,000/kW likely “fishing for investment” and unlikely to be realistic, according to experts. High upfront costs and lower generation capacity could mean many next-gen technologies have inferior unit economics compared to traditional large reactors that have scale advantages. Whilst modular advanced reactor builds could cut typical reactor and site build cost estimates, such modularity benefits are more likely to manifest after the first round of advanced reactor deployments, when developers can fine tune designs and then standardise components. Amongst the experts we speak to however, the outlook is more positive on Molten Salt Reactors (MSRs) due to expectations of lower upfront costs vs large reactors. CAPEX on these reactors could be 40% lower due to these projects having a lower overall physical footprint.
Extending plant lives and brownfield expansions is also an option for delivering projects on time and budget. As an example, an expert consulting with Third Bridge sees the opportunity for Dominion Energy to accommodate an additional 4 GW of capacity through advanced reactor builds on the existing North Anna site, subject to constraints like cooling water and transmission.
Regulating the future
Another significant complication when it comes to rolling out advanced reactors is regulation. Manpower shortages, attrition and lack of experience with non-light-water reactor technologies at the Nuclear Regulatory Commission (NRC) are also likely to further elongate approval timelines in the US. Our experts indicate that the NRC’s decisions on how to regulate the production and transportation of key reactor components and transportation are also likely to be headwinds, potentially hitting return on investment (ROI) forecasts. Despite this, there are some promising indicators that the NRC may trend towards more streamlined approvals. An industry insider we spoke to recently attained a construction license for a molten salt reactor with the permitting process taking two years. All-in-all, advanced reactor permitting could take four years for full approvals (two years for a construction licence and two more for an operating licence) and see the first advanced reactor operating by 2030 at the earliest.
Supply chain constraints
The sudden surge in advanced reactor builds means supply is catching up with demand and supply chains are tight with key components and inputs coming from a concentrated vendor base. As an example, Centrus Energy is the only current producer of high-assay low-enriched uranium (HALEU) that the majority of SMRs will need as fuel, but produces around one tonne per year that goes to the government. Whilst Urenco Group is developing a 10-tonne facility in the UK due for completion in 2031, Third Bridge experts believe that 100 tonnes of HALEU at least is needed to service current production forecasts – especially since SMRs can consume around four or five times more uranium per kg of fuel than light-water reactors.
The SMR outlook
As we can see then, driven by growing demand for data processing capacity, key players in the advanced reactor space are racing to develop commercial-scale advanced reactors with a number of models being trialled, each with relative advantages and disadvantages. Alongside this, players of all shapes and sizes face significant regulatory and supply chain challenges that will also need to be navigated as the industry scales over coming years.
While the feasibility of advanced reactor technologies and which players will be the first to deploy them commercially remains uncertain, there is a clear interest in the subject given concrete moves by hyperscalers to make advanced nuclear a tool in their box of energy solutions.
