Fueling Artificial Intelligence's Energy Demands

Recently, SoftBank Group, OpenAI, and Oracle Group announced they are forming a $100 billion joint venture with a goal of increasing to $500 billion over the next five years to build artificial intelligence (AI) and data center infrastructure. Soon after, the Wall Street Journal reported that Santee Cooper, a major power provider in South Carolina was seeking proposals to restart the construction of the V.C. Summer Nuclear Station to fuel the need to secure energy sources for data centers. This follows announcements in 2024 from tech giants such as Amazon, Microsoft and Meta that they are looking to either restart decommissioned nuclear power plants such as Three-Mile Island or partner with advanced reactor and small modular reactor (SMR) developers to power their growing demands for electricity.

But with all of these announcements, one must ask, is this even possible to execute such ambitious growth targets within America's power industry that has stagnated and heavily regulated for the past two decades? It will surely be a challenge and a major shift in public-private coordination is required to even make these goals feasible. In this blog, we touch on prior periods of rapid growth in electricity demand in the U.S., discuss the immense energy requirements from AI and subsequently, data centers and analyze how the U.S. could meet the energy requirements from AI.

Moving to denser sources of energy: looking back at America's electricity consumption

Throughout America's history, energy consumption has of course exponentially increased, but more importantly, the sources of energy have gone from less-dense to more-dense sources of fuel. Take a look at the chart below that is taken from the U.S. Energy Information Administration (EIA) displaying U.S. energy consumption measured in Btu's and disaggregated by energy source since 1775. Prior to the Industrial Revolution, wood was the preeminent form of energy until the mid to late 19th century after western society began to innovate, requiring denser forms of energy. Coal became the fuel of choice as industrialization and westward expansion required greater demands for energy as manufacturing production increased and railroads were built to carry goods longer distances. It was the economic, reliable and high energy density of coal that enabled American innovation and progress in the late 19th century.

In 1901, near Beaumont, Texas, the Spindletop oil discovery marked the birth of the modern oil industry in the U.S. Here, the Lucas Gusher produced over 100,000 barrels of oil per day, leading to a Texas oil boom and the emergence of major companies like Gulf Oil, Texaco, and ExxonMobil creating excitement over the potential of American petroleum reserves. By the 1910's discoveries in California led to the state producing nearly 22% of global oil production!

As nationalism swept through Europe creating fractures in the global order, WWI erupted and the growing need for oil to power machines and weaponry, the birth of the American oil industry took off. Post-WWI economic recovery led to increasing demand for energy, which was then accelerated by the strategic asset that oil became as the world marched into WWII. Oil was the lifeblood for the Allied forces with the U.S. supplying estimates of 85% of their consumption.

Energy was power.

Following the war, again the economic recovery accelerated the use of oil as did the broader use of automobiles as American's began to settle into 'suburbs.' This is where the company, Sears really catered to their upper-middle class American suburban folks with free on-site parking and a variety of appliances and outdoor equipment that could tickle anyones fancy at the time. Driving to Sears in an internal combustion engine, buying appliances that utilized greater amounts of energy from the grid, all required greater uses of energy. At the end of the day, the Golden Age of Capitalism and economic prosperity was founded on the abundance of cheap and reliable energy.

Fossil fuels continued to serve as the foundation of American power and economic prosperity throughout the second half of the 19th century. New technologies in gas turbines and combined-cycle power plants were more efficient (higher energy output per energy input) in addition to the relatively cleaner alternative that LNG provided making it more common in the U.S.

Nuclear's time began in the 1950s as the U.S. created the Atomic Energy Act of 1954 allowing for private industry to own and operate nuclear power plants as well as technological advancements in reactor designs like the pressurized water reactors (PWRs) and boiling water reactors (BWRs) made nuclear more feasible for commercial deployment. Importantly, the geopolitics became a real threat to America's energy and economic security following another wave of nationalism in the Middle East resulting in the 1970's oil crisis and oil embargo of 1973-1974.

Nuclear was a way for America to secure their economic and energy security.

From the beginning of modern society and industrialization, society at large consistently transitioned from a less-dense to a more-dense source of energy as the laws of physics supported societal progress through new forms of dense energy sources.

Throughout the late 20th century and early 21st century, fears of nuclear accidents following Three Mile Island in 1979 and the technological advancements in fracking allowed America to continue to rely heavily on fossil fuels while maintaining a base load source of power through coal and nuclear energy. While the demise of nuclear energy in the early 21st century is a topic for another blog, American politicians during this time forgot the fundamental laws of physics that power modern society as they were supported with private-sector innovations in fossil fuel extraction and a rich natural endowment of resources.

AI and the insatiable demand for energy

The typical Google search consumes 0.3 to 1.0 watt-hours of energy whereas a query using ChatGPT can consume 10 to 100 times more energy, depending on the query complexity. The computational power for large language models (LLMs) does not even compare to a typical Google search!

For most of us, this is how we first were introduced to AI, through ChatGPT. Today, AI applications are being integrated into most industries, from sifting through massive unstructured health databases to help identify patterns that could provide insight into diseases to using AI to detect anomalies in nuclear fission reactions within a nuclear power plant to help reduce costly maintenance and repair. It is hard to fully grasp the implications that such a technology can have on modern society and often it is normal to feel skeptical of the hype. Only time will tell. However, what we can say is that electricity demands have already exponentially increased with forecasts for future electricity demand to accelerate at rates that we have yet to see in decades.

In June of 2024, the Midcontinent Independent System Operator (MISO), which is regional transmission organization (RTO) responsible for managing the generation and transmission of electricity across much of the northern and central part of the U.S. warned that it could face a 1 GW to 3.7 GW shortfall in the norther and central regions, largely due to increased demands from data centers. There have been several reports from other RTOs and utility companies over the last year on realized electricity demand already accelerating due to the growth of data centers. These are not just long-term forecasts, but near-term warnings due to supply-demand imbalances.

There have been a handful of forecasts projecting the estimated energy demand that data centers will require in the years to come and they all come away with one conclusion: a lot more energy is required!

According to Grid Strategies, a consulting firm that specializes in electricity grid modernization and regulatory strategies, their five-year forward projections for nationwide electricity demand load growth has increased 5-fold since 2022 from 23 GW to 128 GW according to their latest December 2024 report. They cite the main drivers as, "data centers and manufacturing" and note that "current load forecasts may not have caught up with growth" resulting in underestimations of projections. These are enormous numbers for any industry, let alone the electric grid industry, which has stagnated for the past couple decades (see charts above from EIA) at around 1% annual growth rates per year.

Grid Strategies notes, while the utility industry has been accustomed to infrastructure expansion and maintenance with a mere 1% growth rate per year, 3% growth rate projections per year would imply "six times the planning and construction of new generation and transmission capacity." Their report highlights the risks of not being able to meet technological innovation demands with the supply of reliable energy by further stating that, "there are real risks to America’s economic, technological, and geopolitical leadership if the grid can’t keep up with demand."

Another report the Department of Energy's Lawrence Berkeley National Laboratory estimates that data center load growth has "tripled over the past decade and is projected to double or triple by 2028." In 2017, total U.S. electricity demand consumption represented around 1.9%, which has increased to 4.4% by 2023 and projections between 6.7 - 12% by 2028. The graph below shows the compound annual growth rate (CAGR) of electricity consumption of U.S. data centers, which was 18% from 2018-2023 and ranging from 13-27% between 2023 and 2028. The report notes that this growth is in addition to a broader context of demand drivers in the future from electric vehicles, onshoring of manufacturing, hydrogen utilization and the overall electrification of industry and buildings.

Investment bank Goldman Sachs estimated that data centers will increase electricity demand by 160% by 2030. While data centers have indeed become much more efficient, efficiency gains have their limits and engineering can only do so much before absolute energy demands require greater buildout capacity. Goldman states that "conversations with technology companies indicated continued confidence in driving down energy intensity but less confidence in meeting absolute emissions forecasts on account of rising demand" supporting this view that engineering efficiencies have their limits when absolute energy demand is growing so rapidly. By 2030, data centers "will use 8% of U.S. power compared with just 3% in 2022."

Over the past year, some of the best performing stocks have been utility stocks such as Vistra and Constellation, which operate a large fleet of nuclear reactors, in addition to a renewable energy portfolio and LNG. Looking at Constellation's September investor report, they estimate after a decade of low growth in U.S. power demand, electricity demand is expected to grow twice as fast through 2030. Consequently, this is why companies such as Constellation and other utility companies are increasing capex to meet this growing demand as customers of this demand need clean and reliable energy to match the growing needs of innovation.

Nuclear's advantages should allow it to become the main source for future data center power

Nuclear energy is one of the best fuel sources to power the growing needs of artificial intelligence. Importantly, it is likely to be the answer for meeting future demand, if society wants to continue to advance and prosper. This is based on how prior historic periods when society went through rapid technological innovations that I highlighted above, society recognized that in order to do so, more dense energy sources must be used. It would be unusual based off of historic precedents if modern society decided to go from a more energy-dense fuel to a less energy-dense fuel.

Yes, there will be other energy sources other than nuclear to help fill the void while nuclear buildout programs are developed and deployed such as renewables with battery storage, LNG and even coal. However, one must just look at the struggles of current battery storage facilities such as the Moss Landing Energy Storage Facility in California that just caught on fire two weeks ago damaging nearly 80% of the batteries in the plant. As one of the largest lithium-ion battery energy storage systems in the world, inefficiencies and physical constraints still exist and in my opinion, showcase the lack of feasibility for significant increases in battery storage to the scale that a large modernized grid can rely on.

Nuclear energy is a proven technology that has many unique benefits due to its inherent attributes that display why one cannot fight the law of physics.

1. High reliability and base load power

   Data centers operate 24/7 requiring constant and reliable power supply. Unlike wind and solar which have capacity factors around 25% depending on weather, nuclear energy averages 93% capacity rates with higher rates in newer designs.

   
   

2. High energy density

   Nuclear energy has an extremely high energy density, meaning small amounts of uranium fuel can generate enormous amounts of electricity. This makes it efficient for powering the large and growing energy needs of data centers. We do not need a thousands of acres of solar panels and sight-specific areas for favorable conditions for wind panels. Nuclear plants can be built technically anywhere with low levels of land required.

   
   

3. Scalability

   As data center demand grows, nuclear energy can scale efficiently with advanced reactors like Small Modular Reactors (SMRs) and Microreactors, which are suitable for smaller loads or localized power needs near data center clusters. Additionally, many people forget that the world already has a proven technology of nuclear reactor designs that can be deployed today. While the U.S. has had troubles of cost overruns on recent large scale reactors, this is because of a variety of factors that are controllable such as: first-of-a-kind (FOAK) cost challenges, skilled labor (that now has retained skills to improve productivity and reduce costs), and regulatory hurdles (an area of focus for the new NRC Director and Secretary of Energy). These can be fixed and new strategies are already being developed and soon to be implemented (see DOE Pathways to Commercial Liftoff: Advanced Nuclear).

   
   

4. Energy Security

   Nuclear energy reduces dependency on volatile fuel markets like natural gas or the minerals needed in wind and solar. While negligence and government failure has created issues of foreign dependency for the nuclear supply chain, these are issues that are controllable and actively being worked on.

   
   

Nuclear energy's unique ability due to the laws of physics that enable its competitive advantage is in my opinion a reason why it will be a key source of energy for data center growth in the future.

To play devils advocate, there are important factors to consider that challenge this notion. Some have been highlighted above, but the main challenges are the high upfront capital costs, regulatory hurdles and first-of-a-kind challenges of rapid deployment of new technologies.

Higher upfront capital costs are unique to the U.S. as many other countries who have not abandoned their nuclear industry for the last two decades do not face capital issues. These will come down as designs and planning improve. Additionally, financing consortiums that include multiple different stakeholders can help reduce the financing risks to one party and spread them out across other users. This is a key component that the DOE's Pathways to Commercial Liftoff discuss. Regulatory hurdles have burdened the nuclear industry as it takes many years of costly investments to get licenses through the U.S. regulatory bodies. The current administration hopes to remove these unnecessary regulatory burdens and accelerate the licensing process. Lastly, FOAK risks and costs are real with new designs and technologies, but can be mitigated with fleet-wide reactor buildout plans and financing structures that spread the risks out.

Lastly, with the news out that China's DeepSeek has a technology that enables much lower-cost and energy-intense designs of AI, the public has gravitated towards the idea that this would be negative for the future of nuclear energy. There a few arguments against this that should be clarified.

First, DeepSeek has not fully disclosed the actual energy-intensity of their model, nor have they disclosed what chip they actually used. There have been reports that they were able to create a business model that utilizes lower-tier chips and are less energy-intensive. However, in a recent CNBC interview with Scale AI CEO & Founder, Alexandr Wang stated that, "the Chinese have more H-100s than people think...DeepSeek has about 50,000 H-100s that they can't talk about because they are against the export controls that the U.S. has put in place." Therefore, the assumption that DeepSeek has been able to achieve these breakthroughs with lower-tier chips should be interpreted with a grain of salt. Secondly, if we assume that DeepSeek did indeed create a more efficient algorithm requiring less energy, wouldn't this increase efficiency generate greater demand and quicker adoption for AI and thus, data centers? I've seen many people support this notion with support from "Jevon's Paradox," which states that increased efficiency in resource use leads to higher overall consumption of that resource, rather than reducing it. Consequently, it is definitely possible that this revelation generates more demand for AI and data center use and at an accelerated and broader scale.

Lastly, there have been reports that Microsoft and OpenAI are investigating whether data output from OpenAI's technology was stolen in an unauthorized manner by a group linked to DeepSeek. This would have major repercussions for the broader usage of DeepSeek's model. White House AI czar, David Sacks also spoke out claiming that there is evidence that DeepSeek used OpenAI's models to develop its own technology through a technique called distillation. In response, there are have also been reports recently that the White House is looking into additional curbs of Nvidia's chips to China.

In conclusion, the recent events over DeepSeek is too early to tell and are at best skeptical. Even with the assumption that they did indeed develop a more energy-efficient model, it is unlikely that this will drastically reduce the demand for AI and data centers and thus, nuclear energy.

Regarding nuclear energy's growth projections, the investment thesis surrounding nuclear energy did not require AI adoption as countries and companies around the world have already taken an initiative to deploy nuclear reactors and bring existing reactors back online.

As we think about nuclear energy's unique characteristics and the historical examples of society advancing only when they adopt a more efficient and energy-dense fuel one must not forget the laws of physics:

The laws of physics strongly favor nuclear energy over intermittent renewables and liquefied natural gas (LNG) for large-scale electric grids due to its unmatched energy density, reliability, and efficiency. Einstein’s mass-energy equivalence (E = mc²) demonstrates that nuclear fission releases millions of times more energy per unit of fuel than chemical reactions in fossil fuels, allowing a single nuclear plant to generate constant base load power for decades with minimal fuel input. Unlike renewables such as solar and wind, which are constrained by the Second Law of Thermodynamics—requiring vast land, storage, and redundant infrastructure to mitigate intermittency—nuclear reactors operate 24/7, unaffected by weather conditions. Additionally, Faraday’s Law of Electromagnetic Induction ensures that a steady mechanical energy source (like a nuclear-powered steam turbine) is superior for grid stability, avoiding the fluctuations seen in renewables. Compared to LNG, nuclear eliminates reliance on volatile fuel supply chains, reducing geopolitical risks and emissions while preventing waste heat losses inherent in combustion processes. By leveraging the strong nuclear force and controlled chain reactions, nuclear energy delivers unparalleled power density, reliability, and long-term sustainability, making it the most physics-optimized solution for powering modern electric grids.