As the data center industry evolves, so does the need for sustainable energy solutions. Bill Kleyman points to nuclear power as a potential game changer that is gaining traction. I really wanted to write this article. In recent years, the debate about energy consumption in the data center industry has continued to increase. However, in the last year, a substantial change has occurred at a technological level, which has increased the urgency of talking about renewable and clean energy.
Let’s start with the thesis statement: The pace of our technological evolution is rapidly becoming unsustainable.
Consider this statistic from research firm Omdia. In Q2 2023, Omdia estimates that Nvidia shipped more than 900 tons (1.8 million pounds) of H100 compute GPUs for artificial intelligence (AI) and high-performance computing (HPC) use cases. You read that right: we now measure GPU shipments by weight instead of units sold. Anecdotally, at a recent conference, Chris Crosby, CEO of Compass Data Centers, did some rough calculations on this topic. If we put all these GPUs online at once, they would consume more than 30 GW of energy. How can we host this use of a lot of energy? At this moment, which market can return and say that we have so much clean energy to provide cases of use of the AI? In a recent conversation, utilities said they would have to build coal-fired power plants to support these new use cases. It’s a little ridiculous that we have to use fossil fuels to power some of the world’s most advanced technologies. Before we go any further, I don’t want this to be a post about generative AI. So, let’s get a few things straight. We know that this technology will not disappear. What we are experiencing now is not so much a technological change as a change in humanity. We have completely changed the way we interact with data because we can query a dataset for the first time in human history and get an “informed” answer. Even if you haven’t used ChatGPT, you are a generative AI user if you have simply used Google or Bing in the last few months. This evolution has happened rapidly over the past year.
Now that we’ve cleared that up, let’s talk about the 500-pound gorilla that GPT generated in the room. That is, how do we feed all these AI ecosystems that require a lot of density? In AFCOM’s State of the Data Center 2024 report, a majority of respondents (53%) believe that new AI (generative AI) workloads will increase capacity requirements for the colocation industry. As a result, energy requirements will also change.
The amount of energy consumed by generative AI can be quite staggering. A single Google search can light a 100-watt light bulb for about 11 seconds. GPT-type instances can be 600 to 800 times more powerful than a single Google search. Moving to a New Energy Source: Nuclear Energy
As quickly as generative AI has taken over the market, new conversations have emerged about energy sources that have been around for a while. Specifically, nuclear energy. “We envision a world where there are no limits to energy,” says Bret Kugelmass, CEO of Last Energy, a developer of small modular reactors (SMRs) for private-sector entities, including data centers. “We want people to live energy-intensive lifestyles. For this to happen, energy production must be decoupled from environmental impact, and therefore clean energy must be abundant and affordable.”
I discussed this topic with Bret and he raised several key points about what the nuclear industry needs to do to meet the growing demand for data center power. First, it needs to maintain a strong focus on deliverability, starting with the technology. There are many new reactor designs, all of which are interesting from an engineering perspective. However, the winning designs will leverage proven technologies to serve this market and achieve commercial success. Last Energy uses a proven light water reactor design modernized for scalability. By converting the design into a product and adapting it to commercial uses such as data centers, they can deliver results faster, more cost-effectively, and minimize project execution and financial risk. Last Energy SMRs are up to 95% pre-assembled, delivered by truck, and assembled on site. Although each unit is a 20 MW SMR, demand can be met by installing as many as the customer requires rather than building (and licensing) a custom plant design. The formula is paying off, and the company has contracts in place for several units of its PWR-20 power plant.
According to Kugelmass, nuclear power is not only the best option for the industry, it may be the only one: “The data center industry is an ideal application for on-site nuclear power. This is a sector that is growing as fast as development plans and supply chains allow, and is facing increasing energy supply constraints. The industry is committed to achieving carbon-free energy, but it needs the baseload energy traditionally provided by gas or coal-fired power plants. “Nuclear is the answer, and the industry has been quick to embrace it.”
Last Energy isn’t alone in pushing for more nuclear power for our industry. The Nuclear Regulatory Commission has already approved another small modular reactor (SMR) design in the United States: NuScale’s advanced “light water” SMR, which generates more than 400 MWe.
In October 2023, hosting provider Standard Power announced plans to use NuScale SMRs to build two nuclear power plants in Ohio and Pennsylvania to provide nearly 2 GW of power to nearby data centers by 2029. And if we look at the Northern Virginia market, in April 2023, Green Energy Partners, a property and project development firm, purchased 641 acres for a project that calls for using four to six SMRs to power 20 to 30 data centers, generate hydrogen fuel, and provide backup power to the Virginia grid.
And if we look at more recent developments, we already see data center operators paying millions of dollars to have access to nuclear power. Equinix just made a $25 million upfront payment to nuclear reactor company Oklo to purchase up to 500 MW of nuclear power.
Growing Interest
There has been growing interest in nuclear systems in the data center space. Between this year and last year, respondents who said they will use or at least consider nuclear energy have more than doubled to nearly a quarter of respondents (21 percent, up from 10 percent in AFCOM’s State of the Data Centers report last year), according to AFCOM’s State of the Data Centers report.
At this year’s AFCOM Data Center World event, Dr. Rian Bahran, who currently serves as deputy director for nuclear technology and strategy in the White House Office of Science and Technology Policy (OSTP), explored the dynamic changes in clean energy use and the policies that are transforming our world. In his keynote address, Dr. Bahran focused on the critical role of nuclear technologies in strengthening energy networks by providing a stable and efficient energy source that complements intermittent renewables such as solar and wind. A highlight of his talk was the discussion on the intersection of nuclear energy with digital infrastructure. “Last October, the Administration issued a historic executive order to ensure that the United States is at the forefront of harnessing the promise and managing the risks of AI, including at the forefront of innovation and competition,” Bahran said.
“In fact, the AI job market is growing rapidly and currently comprises 2% of all jobs created in the United States. All this technological progress is adding additional burdens to our electricity generation infrastructure and has the potential to have a drastic impact on the energy and electricity sectors.
Bahran added: “Our best estimates suggest that data centers currently consume between 1 and 2% of annual electricity consumption in the United States and that the IT sector as a whole represents about 5% of total U.S. consumption In high-demand computing regions like Atlanta, Northern Virginia, Phoenix, and parts of Texas and California, data centers account for much larger shares of regional loads, putting significant pressure on the grids and utilities that supply them with power. “Many national estimates suggest that data center loads could double by 2030.”
In terms of generative AI and new data applications, Dr. Bahran delved into nuclear energy and how it could impact digital infrastructure. “AI is about 40% of data center load and is the main driver of future growth. While there is clearly uncertainty in these predictions, driven by questions such as the pace of AI adoption, the shape of AI business models, and the potential for future efficiency, we know that combining data centers with nuclear reactors that generate high-quality, well-paying nuclear jobs for decades is a great idea.”
These conversations cannot come soon enough for those working in energy and digital infrastructure.
It’s a chicken-or-egg situation, says Andrew Bochman, a nonresident senior fellow at the Global Energy Center and senior network strategist and infrastructure advocate at Idaho National Laboratory (INL). “We’re halfway through 2024 and we’re entering what some call the ‘electrification challenge.’” “It’s an old, crumbling electric grid that’s going to be asked to support a lot of new workloads that are going to help decarbonize and support the economy.”
But Bochman stresses that this electrification modernization won’t be easy. “And to make matters more difficult, we have to make sure that it happens without adding more CO2 or methane to the atmosphere,” he said. He added some good news: “The answer may lie in an area of physics that we took advantage of in the middle of the last century. And despite the many challenges it has faced in terms of public perception and costs, one thing I’ve been saying for a long time is that when climate concerns eclipse nuclear energy concerns, everyone will want one [an SMR or microreactor].
To drive home the point, Bochman offered some examples of how these nuclear resources will affect the current power limits of AI. “If they were approved by the NRC and could be built at scale today, especially to support the rise of GenAI, we would see them flying off the shelves. Unfortunately, there’s a bit of a delay. But people at my national lab in Idaho, in industry, and elsewhere, are working around the clock to make new designs a reality as quickly and safely as possible.”
Some Final Thoughts – Bill Kleyman
I am very excited about these latest developments. Emerging programs and government initiatives are supporting SMRs and advanced nuclear systems. These programs are at the forefront of nuclear technology and offer safer, more flexible, and cost-effective solutions to our modern energy needs.
While I would love to discuss this topic in more detail, I am limited in what I can include in one article. But I want to touch briefly on safety. When I spoke to Bret from Last Energy, he was clear: “Water reactors, even the old ones, don’t have the inherent risk profile that people think they do.” Furthermore, modern designs have made nuclear technologies much safer than ever before. Finally, to advance AI, we will have to be creative. We are asking data center leaders to deploy physical infrastructure that is much more energy-intensive than it was a year or two ago. The reality is that we need to look for new, clean, and consistent energy sources that can complement intermittent renewables. I believe the rise of artificial intelligence will also lead to the rapid emergence of new and innovative energy solutions to support modern digital infrastructure.
