“Demand for data centers is increasing." We have seen this in the headlines for several years now, and, just like any revolution in human ingenuity, the speed at which technology develops is the deciding factor. The latest evolution of artificial intelligence (AI), which made tools like ChatGPT and Auto Machine Learning (ML) available to consumers, vastly increased the demand for data centers that can manage the enormous amount of power, storage and cooling requirements of AI technology. The rate at which data centers must be developed has been reset, and along with this development comes the question of how to power them.
As per the latest reports, the global demand for data center capacity could rise at an annual rate of 19 to 22 percent from 2023 to 2030 to reach an annual demand of 171 to 219 gigawatts (GW). To put this into perspective, the current demand is 60 GW, which signifies that we need to ramp up the capacity of data center construction at nearly four times the current speed to keep up with the growing demand.
As the semiconductor industry and data center designers meet the hardware needs for data centers by designing faster chips (for example, Nvidia’s GB200 or Google’s Trillium TPU) and rack densification techniques, the ability to power these data centers becomes a critical part of the equation. However, the speed at which utility companies can deliver power is a major concern for data center providers. Current means of power generation and transmission must be enhanced to keep up with the demand. From a global perspective, the number of data centers in the U.S is nearly ten times the number of data centers in Germany or the UK, making the risk of inadequate power supply very significant in the U.S market.
Grappling with grid constraints
Many data center developers had assumed that if they could connect to the grid, their local utilities would find a way to supply the electricity needed, but it is getting harder for utilities to provide what developers are asking for in terms of quantity of power, due to inadequate generation facilities, and quality of power, due to outdated transmission and grid infrastructure. Furthermore, due to substantial investments in the manufacturing sector, critical equipment (such as transformers or cooling modules) and talent resources are in short supply, which can result in extended build and interconnection timelines. Consequently, it is increasingly difficult to get data centers connected to utility power. The realization that the grid cannot provide sufficient power in most places is causing data center developers to consider alternatives to utility power and take steps to lessen or completely remove their reliance on the grid.
Another consideration that is becoming increasingly important to developers is the source of the power, as most major tech customers driving the demand, like Google, Meta or Microsoft, have set goals for net zero carbon emissions across their value chain. The solution is to have reliable, clean power that can scale up at the same pace as the growth of computing power needs—it’s abundantly clear that conventional fossil fuel energy sources are not clean, and we need to focus on renewable and clean power sources such as solar, wind, hydro and nuclear. Based on the current growth trajectory, together with nuclear power, renewable and low-emissions sources are set to generate more than half of the world’s electricity before 2030.
Finding the right source of clean power
While renewable power sources like solar and wind are clean, they are not robust enough for essential data center infrastructure, due to climate and geographical factors. Currently, solar and wind work as supplementary sources of energy for the grid, with fossil fuels and nuclear power serving as the primary baseload supply. Technologies such as distributed energy resources, uninterruptible power supply, and battery energy storage systems fuel cells can be used with solar and wind to create an always-on microgrid, ensuring a constant reliable power supply, but continuous upgrades and scalability concerns can be an ongoing challenge for end users.
As per the U.S. Department of Energy reports, nuclear power generation stands out from other clean energy sources as it remains unaffected by weather variables and boasts the lowest lifecycle greenhouse gas emissions and the lowest ratio of physical area versus power generated. Based on reliability and environmental requirements, nuclear power is the most natural choice for spearheading the scale-up required in power generation.
Nuclear is the right choice, but are we ready to implement it?
While nuclear appears to be the natural choice for data center power generation, its widespread implementation is not without roadblocks.
Current nuclear power sources consist of large-scale power plants that house reactors with a capacity that ranges from 700 to 1,000 megawatts (MW) of power. This is a great option for data centers, however, the construction of large-scale nuclear plants has long been plagued by cost and schedule overruns. The latest plants constructed in the U.S., for example—Vogtle Units 3 and 4 in Waynesboro, Georgia, commissioned in 2023 and 2024—took 15 years to build and cost $36 billion, nearly twice the projected timeline and cost, according to Third Act.
Given the shift in public perception toward nuclear power and the stringent regulatory approval process, along with the implementation of numerous redundancy measures in design and significant capital investment, the scale of traditional nuclear plants does not always align with the principles of economies of scale. The bigger the plant, the higher the derailers.
-
Read more
In response to the hurdles encountered by large reactors, industry experts have been working to downsize reactors to a smaller, more flexible scale. Initiated in the 1960s, the development of a small modular reactor (SMR) involves designing a smaller, modular version of the traditional nuclear reactor. SMRs are advanced nuclear reactors with power capacities that range from 50-300 MW per unit, compared to 700+ MW per unit for conventional nuclear power reactors. Due to their smaller footprint, SMRs can be installed in locations unsuitable for larger nuclear power plants. Modularized SMR units can be manufactured, shipped, and installed on-site, making them more affordable and faster to build than large power reactors. They can also be deployed incrementally to match increasing power demand. Another significant advantage is that these SMRs have lower fuel requirements and can be refueled every three to seven years, compared to every one to two years for conventional nuclear plants, with some SMRs being designed to operate for 30 years without refueling. In the modern SMR space, organizations such as TerraPower, Kairos and X-Energy are in the forefront, with the support of tech giants Microsoft, Google and Amazon respectively.
Even though SMRs have benefits, certain challenges remain. The major challenges among them include maturity of the technology, constrained supply chains, regulations that are not adjusted to the smaller models, and fuel production.
Foremost among these challenges is fuel. More than half of the SMRs currently in development use HALEU (high assay low-enriched uranium) fuel, which ranges from five to 20 percent of uranium-235—beyond the five percent level that powers most nuclear power plants in operation—allowing for more efficient energy production. While HALEU is the best option for fueling SMRs, the material is not yet widely available and limited to research reactors and medical isotope production. For SMRs to remain a viable option for powering data centers, there is clearly a need to increase fuel production.
Another big and costly hurdle for SMR use is obtaining a license. Organizations have spent a substantial amount of investor funds and time in developing standard designs and working through the NRC approval process, only to cancel the projects due to economic viability. The nascency of the technology cannot sustain the regulatory requirements based on older and larger models. Current NRC regulations make it difficult procedurally and administratively for advanced reactor developers to license new reactors that take advantage of small, innovative designs. Hence there is a need to refine the regulatory framework based on current SMR designs.
The AI revolution has accelerated the demand for computing power via data centers, which in turn has created the necessity to produce and transmit clean, reliable power to operate them at an accelerated pace. On the demand side, conditions are favorable for SMRs to take off, however, on the supply side we have seen that early attempts to build commercial SMRs have floundered due to costs and limited fuel technology. Technological challenges seem surmountable, but the economics must work for SMR producers. Nuclear industry experts, regulatory authorities, supply chain networks, the construction industry and communities must collaborate to chart the plan forward.
There is clear demand for AI technology and the resources to power it, and as technology companies are capable of funding the initial risk-taking and scale-up that SMRs require, there is no better time to move forward. SMRs have the potential to revolutionize the needed increase in power production and supply, propelling the AI revolution for the decades ahead.
About our author
Girish Pillai is a program management professional with a background in engineering and law, specializing in EPC planning and project controls management. His extensive experience includes power plants, rail infrastructure, semiconductor facilities, and data center construction projects, where he drives efficiency and successful project execution.
