Energy Future: Powering Tomorrow’s Cleaner World

Power Grab: Data Centers and the Grid Part 3 - Supply Strategies

• Peter Kelly-Detwiler

Share episode
Can data centers sustain their rapid growth without jeopardizing our energy future? In this episode, we promise to uncover the intriguing dynamics of power supply challenges faced by data centers and the innovative solutions needed to address exponentially increasing energy demands. As data centers become ever more essential, they transform into energy-hungry giants, especially during intensive AI training and inference processes. Our discussion highlights collaborations with nuclear power plants, like Amazon's link with Talon Susquehanna and Microsoft's efforts at Three Mile Island, along with groundbreaking alternatives such as modular nuclear plants and geothermal projects. These strategies reveal the intricate landscape of energy procurement, essential to maintain the relentless pace of tech-driven development.

Our exploration also ventures into the emerging trends shaping data center power consumption. Facing mounting pressure to boost computing power, data centers are increasingly reliant on diverse energy sources, including the controversial use of the gas grid, potentially leading to higher carbon footprints. We delve into the vital role of third-party independent power producers in enabling swift expansions, while also addressing the impact of AI workloads on energy prices and market competitiveness. Don't miss the insights on how the bulk power grid might be reshaped by these tech demands and what the future holds for this ever-evolving industry.

Support the show

Speaker 1:

In our first two sessions we looked at the enormous projections for data center growth and the amount of energy that may be required to serve that potential demand. We then discussed issues related to chips, power draw, availability of data and other issues affecting growth in future power demand. In today's piece we'll examine the potential supply options that data center owners and operators are chasing. Later we'll discuss implications for competitive power markets as well as distribution utilities, but those topics are for another day. So let's talk supply. Of all the limiting factors determining data center growth, power is the big one. A steady supply of electrons is necessary to feed these extraordinary machines and the power-hungry racks of chips they host. Let's look at where that energy gets used. Today. Most of it is used for training the computers, crunching through data, to make connections and create intelligence. Training and inference are estimated to consume about 70 to 80 percent of the power used In training. Most of the energy is used in the initial run model, which can take weeks or longer. 20 to 30 percent of processing power use is dedicated to cooling, though highly efficient data centers can get those cooling numbers down into the teens. The amount of energy used for inference, where those decisions are made from the trained models. That's growing rapidly. Inference doesn't have to be performed by energy-thirsty GPUs, but the usage adds up when you ask a question to chat or Perplexity or Gemini or a host of other applications. Those queries burn through about 10 times the energy it would take to run a standard Google search. As we increase our number of queries and I've performed at least five just since I started researching this section then the aggregated usage soars started researching this section. Then the aggregated usage soars, meta, reported in a 2022 paper that 30% of their energy use was devoted to training, including experimentation affecting algorithms and modeling approaches, while inference for its own internal Facebook use represented 70% of total electricity consumption.

Speaker 1:

So if data centers cannot escape the insatiable hunger for power, where will they get it? Well, first, as we've already discussed, they'll grab as much from the power grid as they can, wherever they can. Globally, there's a preference for the US with its large grid, stable economy, rules of law and access to communications cables. Europe would also fit the bill in many ways, but the grid is old and space is at a premium, so some data centers migrate to Malaysia, singapore, brazil anywhere else they can get power. China's its own unique case with geopolitical implications. Indeed, as dominance in AI increasingly becomes a national security issue, geopolitics increasingly enters the fray. Fortune reported in October, for example, that since the release of chat GPT two years ago, the US Department of Defense has awarded $670 million to over 300 companies working on AI-related projects, and AI is already being used to shape the Ukrainian battlefield.

Speaker 1:

Sticking with the US, then, the supply strategies generally look like this First, get as much power as quickly as one can directly from the utilities and the grid. We'll talk more about that in a coming session. Failing that, tie into existing assets and co-locate where possible with them. This is what Amazon Web Services did with Talon Susquehanna Nuclear Plant, where they were able to access 300 megawatts of supply at a co-located data center before the FERC gave them the Heisman and rejected plans for an additional 660 megawatts. It's also what Constellation and Microsoft are playing on with their 20-year 835 megawatt agreement to resuscitate Three Mile Island Unit 1, shuttered since 2019 for economic reasons. Constellation may spend up to $1.6 billion on revitalizing the plant and is seeking a federal loan guarantee to make it happen. Meanwhile, nextera recently announced it's in talks with the feds and conducting engineering assessments as they eye restarting the 600 megawatt Dwayne Arnold nuclear plant in Iowa that was taken out of service in late 2020.

Speaker 1:

There are, however, only so many existing and recently closed nukes to go around. So what then? Well, we may also see a few smaller facilities supplied by fuel cells. In May, bloom Energy and disclosure I own some shares announced it was significantly expanding on its original 6.5 megawatt supply agreement to an Intel data center in Silicon Valley, though it gave no details on size. In June, bloom announced a 15-year 20-megawatt deal with Amazon Web Services for a planned Silicon Valley data center and in its recent earnings call, bloom's CEO teased more. But really, who cares? In the face of the tens of thousands of megawatts being sought by data centers, these quantities are akin to a pimple on a fruit fly in terms of significance to anybody but Bloom to advanced and enhanced geothermal projects and nuclear.

Speaker 1:

Google kicked off the geothermal game with a 3.5 megawatt deal with developer Fervo Energy in Nevada that delivered its first juice in late 2023. On the heels of that, in June of 2024, google committed to buying 115 megawatts of additional 24-7 energy from Fervo under a long-term contract with Nevada utility Envy Energy. Energy from Fervo under a long-term contract with Nevada utility Envy Energy. Then, in August, sage Geosystems stepped into the ring, announcing a 150 megawatt supply agreement with Meta for geothermal from a site quote east of the Rockies, unquote. Other projects will undoubtedly follow as advanced geocompanies get better at drilling, cut costs and continue to improve their technologies. But these first large projects won't come online for a few more years and the industry isn't going to get tens of gigawatts there anytime soon.

Speaker 1:

So what about nukes? Modular nuclear plants, those small ones that can be built in the factory, shipped and assembled on site. Here too, we've seen numerous recent announcements. Small reactor company Oklo inked an April deal with data center co-location company Equinix for 500 megawatts, with start dates to be determined. Google signed an October 2024 deal with startup Kairos Power for 500 megawatts of modular nukes to be online between 2030 and 2035. Between 2030 and 2035. Also in October, amazon Web Services announced it had made an investment in modular reactor company X-Energy as well as signed an agreement with a group of Washington State Utilities, energy Northwest, for 320 megawatts to be furnished by four 80-megawatt reactors, with an option to expand to 960 megawatts. Eventually, delivery date is expected to be in the early to mid-2030s.

Speaker 1:

And that's the rub here the modular nuclear industry won't scale and deliver large quantities to support AI data centers anytime soon. And, with a singular exception of NuScale, none of the numerous players even have design approval from the Nuclear Regulatory Commission. Nimbyism will undoubtedly also be an issue, limiting siting opportunities, while waste security and disposal will also need to be addressed. So where do we then stand? You've got a data center industry desperate for power, willing to pay nearly any price for it, and a remarkably sclerotic grid. If new technologies are to be the cavalry riding to the rescue, don't hold your breath. The cavalry's horses haven't even foaled yet.

Speaker 1:

The logical outcome, then, is to bypass the power grid entirely and go right to the source the gas pipeline. Gas is at historically low prices and fracking continues unabated. Heck. Much of the US gas production is a byproduct of oil drilling. In the prolific Permian Basin, a huge portion of gas production is associated gas from wells aimed at lucrative oil, not even targeting gas. Not surprisingly, data center operators are now looking at building their own generation, co-located with data centers, giving the term co-location a whole new meaning. Numerous gas pipeline companies report discussions with data center operators for direct gas hookups to support behind-the-meter generation where a grid would represent the backup power source, for example. Energy transfer, is said, is in discussions with data centers for new demand in excess of 3 billion cubic feet per day For perspective 2023. Us average daily consumption was 89 BCF per day.

Speaker 1:

The picture is still murky since its early days, but the trend seems quite obvious. It's likely that the data center crowd will get as much as they can from the existing grid, while tapping into the gas grid as quickly as possible, carbon emissions be damned. It's also quite possible that third-party independent power producers will take on the lion's share of this development task. But if the race is really all about speed to power, the urgency to quickly develop new compute capabilities at huge scale this is likely a hashtag. Can't not happen. Type of outcome. All things being equal, new AI load, whether directly supplied or dependent on the grid, will likely have the effect of raising prices. In our next session, we'll talk about the potential impacts on competitive markets and the bulk power grid. Thanks for watching.