I have been looking forward to writing this article. Over the past few years, the discussion around energy consumption in the data center industry has continued to increase. However, over the past year, there has been a substantial shift in technology, which has increased the urgency in the conversation around renewable and clean energy
Let’s start with the thesis statement: The pace of our technological evolution is quickly becoming unsustainable.
Consider this statistic from research firm Omdia. In the second quarter of 2023, Omdia estimated that Nvidia shipped over 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’re now measuring GPU shipments by weight rather than by units sold. Anecdotally, at a recent conference, Chris Crosby, CEO of Compass Data Centers, did some back-of-the-napkin math on this. If you bring all these GPUs online at once, it would consume over 30 GW of power.
How do we accommodate this much energy use? At this very moment, which market can come back and say we have this much clean energy to provide for AI use cases? During a recent conversation, utility providers said they would have to bring up coal-powered energy plants to support these new use cases. This is a little bit ridiculous, where we must leverage fossil fuels to power some of the most advanced technologies in the world.
Before we go on, I don't want to make this an article around generative AI. So, let’s get a few things out of the way. We know this technology is not going anywhere. What we're experiencing now is less of a technology shift and more of a shift in humanity. We have completely changed how we interact with data because we can query a data set for the first time in human history and get a “conscious” answer back. Even if you haven’t used ChatGPT, you are a user of generative AI if you’ve simply used Google or Bing in the last couple of months. That’s how quickly this evolution has happened over the past year.
With that out of the way, let’s talk about the 500-pound GPT-generated gorilla in the room. That is, how do we power all of these AI, density-hungry ecosystems? In the 2024 AFCOM State of the Data Center Report, most respondents (53%) believe new AI workloads (generative AI) will increase capacity requirements for the colocation industry. As a result, power requirements will change as well.
The amount of energy that generative AI consumes can be pretty staggering. A single Google search can power a 100-watt light bulb for about 11 seconds. GPT-like instances can be anywhere from 600 to 800 times more powerful than a single Google search.
Turning to a New Power Source: Nuclear Energy
Just as quickly as generative AI has taken the market, there have been new conversations around power sources that have been around for quite some time. Specifically, nuclear energy. “We imagine a world where there is no limit on 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 high-energy lifestyles. For this to happen, energy production needs to be decoupled from environmental impact and, therefore, clean energy needs to be abundant and inexpensive.”
I discussed this topic with Bret, and he brought up several key points about what the nuclear sector needs to do to address the surging energy demand from data centers. First, it has to maintain a laser focus on deliverability, starting with technology. There are a lot of new reactor designs out there, all compelling from an engineering point of view. However, winning designs will leverage proven technologies to serve this market and have commercial success. Last Energy uses a proven light water reactor design modernized for scale. By productizing the design and right-sizing it to commercial uses like data centers, they can deliver faster, more cost-effectively, and minimize financing and project delivery risk. Last Energy SMRs are up to 95% pre-assembled, delivered on trucks, and assembled onsite. While each unit is a 20 MW SMR, they can meet demand by installing as many as a customer needs rather than building (and licensing) a custom plant design. The formula is finding traction, and the company has contracts in place for several units of its PWR-20 power plant.
A prototype of Last Energy's 20 MWe 'micro-scale' nuclear power plant
According to Kugelmass, nuclear is not only the best way forward for the sector – it may be the only way: “The data center sector is an ideal application for onsite nuclear power. You have an industry that is growing as fast as development schedules and supply chains permit that is increasingly facing power supply constraints. The industry is committed to procuring carbon-free power but needs the baseload power traditionally provided by gas or coal plants. Nuclear is the answer, and the industry has quickly realized this."
Last Energy isn’t the only one leaping to support more nuclear for our industry. The Nuclear Regulatory Commission has already approved another small modular reactor (SMR) design in the US: NuScale’s advanced ‘light-water’ SMR, which generates over 400 MWe.
In October 2023, hosting provider Standard Power announced plans to use NuScale’s 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 company, purchased 641 acres for a project that includes using four to six SMRs to power 20 to 30 data centers, generate hydrogen fuel and provide backup power for Virginia’s grid.
And, if we look at the most recent developments, we already see data center operators paying millions of dollars to gain access to nuclear power. Equinix just made a $25 million pre-payment to nuclear reactor firm Oklo to procure up to 500 MW of nuclear energy.
Growing Interest
There has been growing interest in nuclear systems in the data center space. Between this year and last year, per the AFCOM State of the Data Center report, respondents who have stated that they will utilize or at the very least look at nuclear energy more than doubled to almost a quarter of respondents (21%, up from 10% last year).
This year, at the AFCOM Data Center World event, Dr. Rian Bahran, who currently serves as the assistant director for nuclear technology and strategy in the White House Office of Science and Technology Policy (OSTP), explored the dynamic changes in clean energy usage and the policies reshaping our world.
Dr Rian Bahran speaking at Data Center World 2024
In his keynote, Dr Bahran dove into the critical role of nuclear technologies in bolstering energy grids, providing a stable and efficient power source that complements intermittent renewables like solar and wind. A highlight of his was the discussion on the intersection of nuclear energy with digital infrastructure.
“Just last October, the administration issued a landmark executive order to ensure that the United States leads in seizing the promise and managing the risks of artificial intelligence, including in leading the way in innovation and competition,” Dr Bahran said.
“Indeed, the AI-related job market is growing rapidly, currently comprising two percent of all jobs being added to the United States. All of this technological advancement is adding additional load to our electricity-generating infrastructure and has the potential to dramatically impact the energy and electricity sectors.
Dr Bahran added: “Our best estimates suggest that data centers currently consume around one to two percent of annual US electricity consumption, with the broader IT sector representing about five percent of total US consumption. In computing hotspots like Atlanta, Northern Virginia, Phoenix, and parts of Texas and California, data centers represent much higher proportions of regional loads, placing big pressures on the grids and utilities providing them power. Many national estimates suggest that data center loads could double by 2030.”
Regarding generative AI and new applications around data, Dr Bahran dove into nuclear energy and how it might impact digital infrastructure. “AI represents about 40% of the data center load as the major driver of future growth. While there’s clearly uncertainty in these forecasts driven by things like the pace of AI adoption, the form of AI business models, and the potential for future efficiencies, we know that pairing data centers with nuclear reactors that bring high-quality, and high-paying jobs for decades is a great idea.”
These conversations can’t come soon enough for those in the energy and digital infrastructure space.
It’s a ‘which comes first, the chicken or the egg situation,’ says Andrew Bochman, a non-resident senior fellow for the Global Energy Center and the senior grid strategist and infrastructure defender at the Idaho National Laboratory (INL). “We are in mid-2024 and entering what some call the ‘electrification gauntlet.’ It consists of a creaky old electric grid about to be asked to support a motherload of new workloads intended to assist in decarbonizing and shoring up the economy.”
However, Bochman points out that this electrification modernization won’t be easy. "And to make things even more difficult, we’ve got to make this happen while not adding more CO2 or methane to the atmosphere,” he said.
He did add some good news: “The answer may lie in a realm of physics we harnessed in the middle of last century. And despite the many challenges it has faced in terms of public perception and costs, one thing I've long said is that when climate concerns eclipse nuclear energy concerns, everyone will want one [an SMR or microreactor]."
To hammer the point home, Bochman offered a few examples of how these nuclear resources will impact the current AI power constraints. “If they were NRC-approved and capable of being built at scale today, especially to support the GenAI boom, you’d see them flying off the shelves. Unfortunately, there’s some lag time involved. But folks at my national lab in Idaho, in industry, and elsewhere are working around the clock to bring the new designs to fruition as quickly and safely as possible.”
Some Final Thoughts – Bill Kleyman
I’m really excited about these latest developments. Emerging programs and government initiatives support SMRs and advanced nuclear systems. These programs are at the cutting-edge of nuclear technology, offering safer, more flexible, cost-effective solutions that align with our modern energy needs.
While I'd love to discuss this topic in more detail, I'm limited in how much I can fit into one article. I would, however, like to touch on safety briefly. When I spoke to Bret from Last Energy, he clearly stated: “Water-based reactors, even the old ones, do not carry the inherent risk profile that people think they do.” Further, modern designs have made nuclear technologies far safer than ever before.
Finally, to energize AI, we will need to get creative. We’re asking data center leaders to deploy physical infrastructure at vast multiples in energy consumption than they have just a year or two ago. The simple reality is that we must look at new, consistent, clean energy sources that can supplement intermittent renewables. I feel the rise of AI will also see the rapid rise of new and innovative power solutions to support modern digital infrastructure.
