MENU
Powering hyperscale data centers with clean nuclear energy.
As industries and governments recognize the growing energy needs of data centers, leaders in the information and communications technology (ICT) space are transitioning to hyperscale data centers: super-efficient facilities with organized, uniform computing architecture that scales up to hundreds of thousands of servers.
While many of these data centers are major purchasers of renewable energy, ICT experts warn that even hyperscale data centers are facing “the limits of shrinkage,” despite growing dependence on renewables and efficiency gains within the past decade. Data centers of all kinds operate at all hours of the year, and therefore rely on tremendous amounts of energy to sustain computer systems and associated components.
At some point, intermittent renewables will be unable to support the increases in internet traffic and data loads of hyperscale data centers while maintaining efficiency gains. Ultimately, nuclear—an always-on, carbon-free electricity source that can support significant energy demand—is required to meet the rise in global demand for digital services, and to get on track with the Net Zero Emissions by 2050 Scenario.
From the standpoint of a nuclear energy provider, consider the trajectory of hyperscale data centers and how nuclear, a 100% clean energy source, can fuel more efficient and cost-effective facilities for data storage.
While some industries may break down types of data centers based on size, others categorize these facilities based on the type of business model used.
Applying the business model framework, we’re left with hyperscale facilities and two other types of data centers:
%20(1).png)
For our purposes, “hyperscale” is a description of size, while a “hyperscaler” is an enterprise user or another customer large enough to operate a hyperscale facility.
Some of today’s most notable hyperscalers include Google, Facebook, Amazon, and other global ICT companies, which maintain some of the most advanced hyperscale facilities. As owners of the largest and most energy-intensive data centers, several hyperscalers have announced goals and investments to significantly reduce emissions and improve energy efficiency.
Primary examples include Google, which aims to operate its data centers on 24/7, year-round, carbon-free energy by 2030, and Microsoft, which recently signed an agreement to partially power one of its data centers with nuclear energy, supplemented by nearly 100% carbon-free electricity at all other times.
The rise of hyperscalers, and large-scale data centers to meet their needs, has prompted an influx of these hyper-efficient information factories, coined the “hyperscale shift.”
The development of small modular reactors (SMRs) coincides with the hyperscale shift, as well as a shift in the relationship between data centers, carbon emissions, and the environmental responsibilities of ICT companies.
Expectedly, the largest of these companies require substantial energy to operate their data centers. According to the International Energy Association (IEA), combined electricity use by Amazon, Microsoft, Google, and Meta more than doubled between 2017 and 2021—rising to 71 terawatt hours (TWh), or enough to power over six million homes—as their data centers contended with rapidly-growing workloads.
The IEA predicts that overall energy use by data centers will continue to grow moderately over the next few years, and that long-term trends in their energy consumption remain unclear. In the years to come, nuclear can combat this uncertainty with clean, always-on energy, allowing hyperscalers to source and match zero-carbon electricity on an hourly basis.
Although hyperscale data centers are often associated with efficiency improvements, these large facilities still require massive amounts of energy to run 24 hours a day, 365 days a year.
Exact figures on hyperscalers’ energy usage are difficult to obtain, but some sources estimate that an average hyperscale data center uses 20-50 megawatt hours of electricity (MWe) annually: enough to power up to 37,000 homes. McKinsey reports that a hyperscale data center can use the same amount of power as 80,000 households. Given the sheer size and energy demands of these facilities, their electricity usage and subsequent environmental footprints are not fully offset by efficiency gains.
Environmentally, the impacts of hyperscale data centers can vary widely depending on their size, geographic location, water use for liquid cooling and electricity generation, and energy sourcing. Many source power from “dirty” electrical grids, which run on fossil fuels as opposed to cleaner energy sources, such as renewables or nuclear.
For hyperscale facilities as well as other data centers, achieving independence from the electric grid’s cost, reliability, capacity, and carbon footprint will require a truly carbon-free power source.
In response to these environmental concerns and growing energy demands, some global ICT players—including Google, Microsoft, Meta, and Amazon—have made public strides toward reducing carbon emissions, prioritizing renewable energy, and optimizing the efficiency of their hyperscale facilities.
Their data centers are only as efficient as their processors, which can be improved – but only to a point. These improvements must also occur in a fast-changing landscape of digital technologies, including artificial intelligence (AI) and machine learning (ML) models.
The IEA predicts that the combined increase in computing demand, AI, ML models, and other emerging services and technologies will eventually outpace the current efficiency gains in data centers.
Rather than cut back on global demands for data, which are only projected to increase, an alternative option is to look beyond the constraints of efficiency—and the intermittency of renewables—toward the consistency of nuclear power.
To date, the U.S. offers the primary example of a nuclear-powered data center. In January 2023, Cumulus Data announced the completion of the powered shell for its first 48 MW data center, powered by a direct connection to the existing 2.5 gigawatts (GW) Susquehanna nuclear power station in northeast Pennsylvania.
Although this facility is not explicitly distinguished as hyperscale, nuclear-powered hyperscale facilities are on the horizon within the next decade. In Virginia, for instance, Green Energy Partners recently announced its plan to create a 1 GW hyperscale data center, primarily powered by two pressurized water reactors (PWRs) in the neighboring Surry County.
These projects offer inspiration far beyond the U.S. As data centers expand in size and computing capacity, their electricity consumption could strain regional power grids around the world, particularly in smaller countries with expanding data center markets. In Denmark, for example, data center electricity consumption is projected to increase from less than 1% to 15% of the country’s electricity use by 2030, based on a 2019 report by the Danish Energy Agency. Left unaddressed, surges in data centers’ power usage could hinder Denmark’s goal of reducing CO2 emissions by 70% by 2030, as well as similar climate goals in other countries.
Compared to both fossil fuels and renewables, nuclear energy is a clean, reliable baseload power with the flexibility to deliver energy wherever it’s needed, independent of weather conditions and time of day.
At the design level, SMRs are especially well-suited for the needs of hyperscale data centers. When constructed with factory-made components, SMRs can be delivered and assembled quickly on-site, and deployed at scale to meet the varied demands of data center campuses.
Last Energy champions this approach with the PWR-20: a 20 MW, factory-built micro modular reactor that can be delivered in approximately 24 months. The PWR-20 supplies carbon-free, always-on energy to satisfy the power requirements of hyperscale data centers and can be deployed incrementally to match increasing energy demand.
Some companies commit to “100% renewable energy matching” by purchasing renewable energy to fulfill their annual energy needs. However, this phrase can be misleading, as renewable energy infrastructures cannot operate 100% of the time to achieve full decarbonization.
Traditionally, when renewable energy sources like wind and solar are unavailable, data centers must instead rely on electricity from fossil fuels. SMRs like the PWR-20 resolve this dilemma by supplying reliable, baseload power during times of high demand, even when that demand exceeds the available wind or solar power.
.png)
Using power purchase agreements, Last Energy provides hyperscaler facilities with the appropriate number of PWR-20 units to support their energy demand. As demand rises, hyperscalers can increase the number of units as needed to fulfill that demand with increased energy efficiency.
If more data centers utilized nuclear energy supplied by SMRs or microreactors like the PWR-20, we may not need to restrict or modify our current data diets, instead entering an era of “energy abundance.” In this era, we’d generate vastly more energy than we currently use and make it zero-carbon, affordable, and accessible.
Transitioning to energy abundance requires a broader recognition of the value of nuclear and of the contributions of hyperscale data centers to global carbon emissions. With both recognition and action on behalf of hyperscalers, the “hyperscale shift” can incite a parallel shift to carbon-free nuclear energy.
%20(1).jpg)
%20(1).jpg)
