Will power-hungry data centers overwhelm the grid? With Fredrik Ellekjær and Victoria Fethke

Let's Talk Energy – Transcript Episode: Will power-hungry data centers overwhelm the grid? Host: Noah Brenner | Guests: Fredrik Ellekjær, Head of New Energies, Rystad Energy & Victoria Fethke, Lead Data Center Analyst, Rystad Energy [0:00] Welcome to Let's Talk Energy, your go-to podcast for smart energy insights. I'm Noah Brenner, and each week we bring you an inside look at the dynamics shaping global energy markets through in-depth conversations with our own Rystad Energy experts as well as some special guests. Data centers have been around since the advent of the internet, but their buildout and associated energy needs have exploded over the past few years, driven by the race to develop ever greater artificial intelligence capabilities. Governments are pushing AI progress and sovereignty over data as the new front in the intensifying global industrial competition. And so-called hyperscalers are pouring hundreds of billions of dollars into data center development, particularly in the US. However, fears of astronomical power demand swamping legacy grids and pushing up consumer prices, as well as general uncertainty over the pace and scope of the AI revolution, are complicating the outlook for the data center buildout. So just how much power will data centers really consume over the next 5 or 10 years? Where is that power most likely to come from? How much could data center developers spend relative to oil and gas or power companies in the years ahead? How might this all impact consumer prices? And can data centers coexist with the grid, or are we more likely to see them put their own dedicated power sources behind the meter? We're going to try to answer all of these questions and more by breaking it down into five key numbers you need to know to understand the intersection of data centers and energy. To help us do that, we have Fredrik Ellekjær, Head of New Energies for Rystad Energy, who works in our renewables, power markets, and clean tech research, joining us from Oslo. And from New York, we have Victoria Fethke, Rystad Energy's lead data center analyst. Fredrik and Victoria, welcome to the program. Fredrik Ellekjær: Hi, Noah. Pleasure to be here. Victoria Fethke: Hi, I'm glad to be here. Noah Brenner: Before we get into the details of these five key impacts of data centers, let's set the scene. Why are energy analysts so focused on data centers right now? Fredrik Ellekjær: Data centers are changing power markets, and so all the energy companies, renewable developers, and power markets care about that. And for the data center operators and hyperscalers, energy is on the critical path for them to build out the data centers. Victoria Fethke: At Rystad we cover the broader value chain of data centers and what goes into building them, and it's become clear that energy is the key enabler that will decide which projects materialize. There are of course other inputs — construction, OEM equipment, compute infrastructure — but it really is energy in the near term that will decide the capacity that is needed. Noah Brenner: Our first number is that global data center demand is expected to hit about 1,250 terawatt hours globally by 2030 — roughly 10% of US power demand. Fredrik, where are we seeing the most intense data center development and what's driving it? Fredrik Ellekjær: The short answer is the US. We currently have 60 gigawatts in the US and by 2030 we're heading towards 120 gigawatts — that's Rystad Energy's perspective. Announced projects are significantly higher than that, and the US represents about half the global buildout. Other regions — China is next, followed by Europe — but there's also a lot of exciting emerging activity in Asia and the Middle East. Zooming into the US, it's a neck-and-neck race between Texas and Virginia, followed by emerging secondary markets including Ohio, Louisiana, Arizona, and Georgia. Virginia is today the largest data center hub in the world and we see continued growth there. But power has become such a constraint that interconnection timelines are stretching up to seven years, so developers are racing to new markets — primarily Texas, where there is available power. Training workloads in particular can be located further from population hubs given their lower latency requirements compared to cloud and inference applications. [4:00] Noah Brenner: Is the US dominance in data center development changing as countries start to think about sovereignty over their data — in the same way they think about sovereignty over energy supplies? Fredrik Ellekjær: That's a very well-put comparison, and even more so in today's geopolitical climate. There are regulatory moves dictating the placement of data centers. In the EU there are now requirements to run on EU-controlled infrastructure. There are localization laws in India and Indonesia. And in the Middle East, where tensions are high, data centers are being built out as sovereign assets owned by state enterprises. That is a shift — not entirely expected — but it is moving away from pure economic logic. Given that power and lead times are not a constraint in some of these regions, there are clear benefits to building in the US, but these factors are offsetting some of those benefits. Noah Brenner: Victoria, what's happening with data center development in China? The US often frames AI as a race with China for industrial competitiveness — does the data center buildout look different there? Victoria Fethke: China is set to become the second-largest market after the US. We are seeing this AI race to build capacity, but the way it's happening is quite different. The first major difference is that US export controls on leading GPU chips needed for training mean China is much more constrained in state-of-the-art training. We're seeing data centers more on the side of inference and application scaling, so China is trailing in terms of the pace at which it can scale large campuses. The buildout is also much more state-directed. A lot of campuses are being pushed inward — for example, to Inner Mongolia where there is available power — while cities are capping large campuses. And the power system has a very different context: the US has had essentially flat power demand for the past 20 years, but China has continuously added new generation and transmission, so it has more headroom and experience to serve this surge in demand. Noah Brenner: Fredrik, how will data centers transform energy markets? Fredrik Ellekjær: Energy markets are already on an electrification journey — changing the way we consume energy from molecules to electrons — and that's been growing at around 0.6% annually over the last 5 to 10 years in the US. Then suddenly data centers come in with this additional demand wedge. If we take everything announced in the US, we will be growing power demand from 0.6% annually to 2% annually. That's a four-fold increase in annual additions. It doesn't sound like a lot of growth, but that's four times the amount of energy that needs to be added to what is already a very large, established energy system. The best way to contextualize it for those in conventional energy: it's the equivalent of growing oil supply from 1 million barrels per day to 4 million barrels per day. Show me any basin globally that can do that. It won't have a huge impact everywhere — not all markets will be equally affected. But to give some of the hotspot numbers: Ireland is looking at a 40% increase in demand from data centers by 2030. Hong Kong at 27%, Finland at 17%, the US at 15%, Singapore at 14%, Denmark at 13%. Ireland's unrisked data center pipeline actually represents 70% of its current electricity demand — that's a massive challenge for regulators and renewable developers in what is a fairly ring-fenced power market with limited interconnection. [8:00] Victoria Fethke: As Fredrik mentioned, the impact on the power market is immense. Two major themes are particularly interesting to follow this year. The first is behind-the-meter power. As data centers race to access power and grid connection times extend to over five years in some markets, operators are procuring on-site power — primarily gas generation, whether gas turbines or reciprocating engines, plus some batteries to smooth out workload demand. We're increasingly seeing players directly forgo a grid connection once they've procured sufficient gas generation. The second theme is hyperscalers becoming much more active participants in power markets — not just as consumers but shaping the grid around them. Google has acquired renewable developer Intersect Power and formed deals with AES to bring renewables and storage to campuses in Texas and Virginia. Microsoft is powering nuclear in Pennsylvania. Amazon is bringing over 3 gigawatts of gas, transmission, and battery to Indiana. We're seeing the convergence of the roles that tech companies play in energy. Noah Brenner: That brings us perfectly to our second key number: roughly 20% of data centers will not be connected to the grid. Victoria, what will drive that number, and could it go lower or higher? Victoria Fethke: The core driver is the time to connect to the grid. If you're looking at more than three, five, or seven years in some markets, players are adopting on-site power strategies — what originally started as bridge power while waiting for a grid connection. But we're increasingly seeing players forgo a grid connection entirely once they have sufficient gas generation. Noah Brenner: Which energy technologies win the race to supply data center growth? Is this a gas-and-renewables-plus-battery competition, or is there a meaningful role for nuclear SMRs, advanced geothermal, or fuel cells? Fredrik Ellekjær: It's all about lead times. You're either waiting five to seven years in the interconnection queue, or waiting the same amount of time for a combined cycle gas turbine delivered from Japan. The energy infrastructure buildout is simply not keeping pace with what the hyperscalers want. That opens up space for other energy sources to help bridge the gap. But big picture, towards 2030 it's still very much a gas-and-renewables-plus-battery solution. We'll increase gas utilization on existing turbines to meet data center demand, while longer-term movements include fuel cells — which now have lead times of months rather than years — and geothermal, which is a highly reliable energy source that data center operators value. Google has a flagship project with Fervo in Nevada. And past 2030, SMRs become increasingly interesting to watch. Victoria Fethke: To add on gas — last year was the first time CCGTs entered the interconnection queue in the US in a decade, showing real interest in building out more gas capacity. But with huge lead times, the ability to add new gas capacity is limited. So we're looking at ramping existing gas plants while adding some capacity from players who procured in advance. [12:00] Noah Brenner: Can data centers be developed in a way that is grid-friendly, and is it actually happening? Victoria Fethke: The grid-friendly data center has come into focus as hyperscalers are increasingly under pressure to ensure fair cost allocation and efficient use of infrastructure. We see this play out in calls for players to pay for incremental grid investment — transmission buildout, substations, local upgrades — as well as bringing new generation to the grid they're entering, which has been termed the "bring your own power" strategy. There's also growing interest in flexibility — essentially data centers not pulling power from the grid during the most expensive hours, so that headroom can be used more efficiently by more players. This is more of a patchwork than a consistent strategy at the moment, but it's an important trend — especially since the White House announced a voluntary commitment where seven major players, including the hyperscalers, OpenAI, Oracle, and xAI, have already signed on. Even though it's voluntary, it shows willingness to be a friendly grid citizen. Noah Brenner: Our next number: in 2026, the world is going to spend more on data centers — roughly $700 billion — than it does on drilling for oil and gas at roughly $600 billion. Hyperscalers alone have announced capex of over $600 billion. We already surpassed solar investments back in 2023. Fredrik, how does this growth cycle compare to other infrastructure buildouts? Is this a historic magnitude? Fredrik Ellekjær: I don't think we've seen anything quite like this — combining double-digit annual growth with significant magnitude and the caliber of players backing it. We're quite rigorous about the 2026 figure — we're monitoring this building by building, by satellite. And it's our belief that in terms of energy infrastructure, data centers are now surpassing what used to be number one: the upstream oil and gas industry, dominated by companies like ExxonMobil. That's quite a pivotal moment and speaks to the magnitude of the industrial buildout we're witnessing. What's also underappreciated is that when hyperscalers guide for $600 billion in total capex, with that comes associated investment into the grid, fiber, water, and OEM purchases. It's a trickle-down effect into the global economy that is substantial. Noah Brenner: What distinguishes the purchasing behavior of a data center operator versus an oil and gas or renewable operator? Fredrik Ellekjær: In order to win this, you need to bring on more data center power than your competitors, and lead time trumps everything. The implication for a renewable developer or oil and gas operator is that they need to think: where do I have overlapping procurement with data center operators? If hyperscalers care only about lead time and less about price, will I be bid out of power purchase agreements? What about rotating equipment — if we're talking behind-the-meter gas turbines — or electricians, truckers, EPC firms? Things become localized fairly quickly, and some of the same EPC firms building out data centers are also building out land-based energy projects. [16:00] Noah Brenner: We would say that data centers are responsible for roughly 0% of US power price increases today — but that will change. Victoria, how is it that data centers haven't really impacted consumer prices yet, and when might we feel that pinch? Victoria Fethke: Data centers haven't been the driver of prices yet. It's mostly been due to aging grid infrastructure, fuel costs, and construction inflation. But this is starting to change, particularly in markets where data centers are coming in near term, such as PJM. In the near term, the effect is being felt through capacity markets — mechanisms that pay power plants to stay online in the future, informed by large load forecasts. We've already seen these prices spike, and in the near term customers are feeling that on their bills. But the bigger long-term story is grid infrastructure buildout — extra high-voltage transmission lines and local upgrades — and the large projects will be more in the 2030s. That is where the consumer impact will be more substantially felt, if cost allocation stays the same. Noah Brenner: So it's as much about the infrastructure needed to service these large loads as it is about competition for power itself? Victoria Fethke: Right. In the capacity market, it's essentially the insurance premium paid for plants to stay online — competing to keep existing generation on the grid and to make it competitive for new generation to come online. Noah Brenner: I want to talk about the emissions impact. We think energy emissions from data centers are likely to increase from around 200 million tons in 2023 up to 650 million tons by 2035. Fredrik, how are big tech companies reconciling their need for power with their climate ambitions? Fredrik Ellekjær: If you source from the grid, that comes with an emissions impact based on the power mix. If you have a behind-the-meter gas turbine, you're burning methane. There is one project with a CCS capture stack to capture the CO2 from a behind-the-meter gas turbine, but that's the exception. The hyperscalers have been on a mission toward net zero — signing renewable energy certificates, green PPAs, buying carbon credits in voluntary carbon markets, and being the largest proponents of facilitating direct air capture and bio-energy CCS. What's happening now is that when they laid those strategies in 2022 and 2023, they may have seen the advent of AI and had some projections, but possibly not to the scale we have now. The implication is that the wedge of carbon removals needed to reach net zero — whether in 2030 or 2035 — has grown significantly. The future of where that's heading is very exciting. [20:00] Noah Brenner: So data centers could be one of the things that spurs carbon markets into the future? Fredrik Ellekjær: It's almost binary. Either we buy a large volume of carbon credits to account for those emissions, or we go back on our targets. And on the first route, it's either we maintain high quality — which is fantastic news for direct air capture and bio-energy CCS — or we reduce quality. Carbon markets are a big mix, from nature-based credits on reforestation to tech removal like direct air capture and bio-CCS. To draw an interesting example of how the data center buildout is impacting other energy sectors: look at Europe's CCS projects that recently reached FID. They would never have reached FID without Microsoft purchases into bio-energy CCS in Denmark and Stockholm. So there are ramifications well outside US borders from how these emissions are being addressed. What I find compelling about carbon markets is that the climate doesn't care where you choose to decarbonize globally, and that enables investment in additional sectors. Victoria Fethke: It's also really driven by the speed at which we can build out the grid, because that impacts how quickly we can work through interconnection queues where we do have a lot of renewables and storage that could play a substantial role in meeting capacity needs, alongside some gas. But if the grid response is delayed, we will see an increased share of behind-the-meter gas. This really requires collaboration between developers, regulators, and data center operators to ensure the grid is built out efficiently. Noah Brenner: Finally, I feel we need to address talk of a bubble in data centers. That's going to come down to what percentage of data centers actually get built and over what period. How are we thinking about risking the global pipeline? Fredrik Ellekjær: Let's talk about the 2030 pipeline, where the announced projects take us — with more projects yet to be announced on top of that. We risk this on an asset-by-asset basis, building by building. The four components to our risking framework are: power availability — which is absolutely at the crux of it, and probably what we've spent 80% of this podcast discussing; offtake quality, evaluating the business model and balance sheet of the offtaker connected to a specific data center; operator quality, looking at the financial stability and robustness of the players building it out; and finally regulatory concerns. Overall, across roughly 13,000 data centers and 8,000 campuses, we're risking the pipeline at approximately 50% — which is quite substantial. Noah Brenner: Is the current rate of growth sustainable? Fredrik Ellekjær: I used to work in advisory and there was a rule of thumb: if you have more than 20% growth over 5 years, look at your forecast again. That holds true because you get unwanted value chain bottlenecks that hamper long-term growth. Fundamentally, it comes back to power availability. We think there's a threshold around 7 to 10% of global power demand in the 2040s to 2050s that represents some sort of potential ceiling — but that all depends on how much compute can be done on devices and whether we start placing data center assets in novel locations, including space. [24:00] Noah Brenner: Is it possible for AI to be a bubble while data centers are not necessarily? Fredrik Ellekjær: The question is less whether AI as a whole is a bubble and more about how fast consumers adopt it broadly — that might see a delay or a reduction in expectations. But in terms of the data center buildout itself, it's very important to note that the hyperscalers have very healthy balance sheets. This is not driven by debt alone but by very high profit margins. That distinguishes it from past infrastructure buildouts. Moreover, the deals already in place are long-term — from building owner to data center operator, you're looking at 10 to 15-year contracts, the kind of longevity infrastructure investors value. From chip owners, there are typically four to six-year offtake contracts. There is some robustness in how this value chain has been contracted up that provides protection against a sudden collapse. Victoria Fethke: We will see data centers around in this buildout. It will be very interesting to see how fast consumer usage speeds up compared to expectations. Noah Brenner: Any parting thoughts or things we haven't covered? Victoria Fethke: Because demand is so high, constraints will appear in value chains outside of energy. Compute is the highest other concern — access to chips and being able to expand manufacturing capacity of leading-edge chips from TSMC, as well as access to the latest generation of GPUs. The time it takes to expand manufacturing capacity in this very concentrated semiconductor supply chain — EUV lithography, for example, is concentrated in one player, ASML — means constraints on one piece can have knock-on effects. While this isn't a near-term constraint, since a lot of compute has been procured in advance, it will be very important to continue following. Noah Brenner: Thank you both for an excellent conversation. Really appreciate it. Fredrik Ellekjær / Victoria Fethke: Thank you for having us. [28:00] Noah Brenner: Let's recap. Data centers will consume growing amounts of power in core development areas within the US, China, and parts of Europe, but we also see that buildout expanding as countries seek more control over their digital infrastructure — much like they think about energy security today. Constraints in energy access and infrastructure could push up consumer prices and move some data centers behind the meter, but a combination of the right policies and planning could limit the impact that data centers have on power pricing. Finally, data centers could make it more difficult for countries and companies to hit their climate goals, but they could also enable investment in the infrastructure needed for electrification and decarbonization. Thanks for listening to Let's Talk Energy. This podcast is a production of Rystad Energy, produced by Lara Rodriguez Scow and Bota Oak. Check out the show notes for further analysis on the topics discussed in this episode, and find us on social media — we're Rystad Energy on all your major platforms. Please leave us a review, subscribe, and hit that like button. You can also keep up to date on our website. If you'd like to send us a question or reflect on today's show, email us directly at podcast@energy.com. Don't forget to join us next week for more Let's Talk Energy.