Latest / Energy Economics with Fexingo: Oil Prices, Renewables, and the Cost of Power / How AI Data Centers Are Reshaping Natural Gas Demand
Transcript
- Lucas: So you've heard the narrative that AI is going to save the world or destroy it, depending on who you ask. But there's a concrete, measurable consequence that utility executives are losing sleep over right now in July 2026: the power demand from AI data centers. Luna: And that's driving natural gas demand, which we're seeing in the numbers. Henry Hub is at $2.83 per MMBtu this week, up from $2.73 last week. Lucas: Exactly. It's not a huge move yet, but the trend is clear. Analysts at Goldman Sachs recently estimated that US power demand from data centers could triple by 2030, with AI workloads accounting for the bulk of that growth. And the thing is, renewables alone can't fill the gap fast enough. Luna: Because you need baseload power that's available 24/7, not just when the sun shines or the wind blows. And natural gas is the cheapest, most scalable bridge fuel we have right now. Lucas: Right. And that's creating a fascinating tension. On one hand, you have tech companies like Google, Microsoft, and Amazon all pledging to be carbon-neutral or even carbon-negative. On the other hand, they're signing power purchase agreements for gas-fired plants to keep their data centers running. Luna: It's almost like a 'do as I say, not as I do' situation. But let's be real: the grid can't handle a 500-megawatt data center popping up in rural Virginia without some serious new generation capacity. Lucas: And 500 megawatts is not an outlier. The largest AI training clusters are approaching that scale. To put it in perspective, that's roughly the same electricity consumption as a small steel mill or about 400,000 homes. And these facilities run flat out, 24 hours a day. Luna: So where is that power coming from? We talked about natural gas, but what about nuclear? Or grid-scale batteries? Lucas: Nuclear is a possibility, but new plants take a decade or more to license and build. Small modular reactors are still years away from commercial deployment. Batteries can smooth out intermittent renewables, but they don't provide sustained baseload for days on end. Luna: So natural gas is really the only option that can scale quickly. And we're seeing that play out in places like ERCOT, the Texas grid operator, which is scrambling to add 10 gigawatts of new gas-fired capacity by 2028. Lucas: That's a massive build-out. And it's not just ERCOT. PJM Interconnection, which covers the Mid-Atlantic, is seeing a wave of interconnection requests from data center developers, many of them paired with gas plants. The queue is backlogged for years. Luna: Let's talk about the economics for a second. When you look at utilities like NextEra Energy — ticker symbol n e e — the stock is up only half a percent over the past five days, at $88.80. But the long-term growth thesis is pretty clear: they own a huge portfolio of gas pipelines and renewable assets. Lucas: And NextEra is actually interesting because they're one of the largest developers of both solar and gas. They're hedging their bets. But the market is pricing in a lot of uncertainty — the clean energy ETF ICLN is down 1.4 percent in the last five days, while the oil and gas exploration ETF XOP is up 3 percent. Luna: Investors are basically saying, 'Show me the demand, and I'll believe it.' And the demand is there. The question is whether the infrastructure can keep up. Lucas: And that infrastructure is expensive. Building a new combined-cycle gas plant costs around $1 billion per gigawatt. And that's before you account for the pipelines to bring the gas in. So we're talking tens of billions of dollars of investment over the next five years, just for data center power. Luna: That's a huge capital commitment. And it has implications for the broader energy transition. If we're locking in gas-fired generation for another 30 years, that makes it harder to hit climate targets. Lucas: Yeah, it's a real dilemma. And I think the honest answer is that we're going to need both: gas for reliability, and renewables for the long haul. But the speed of AI adoption is forcing a near-term compromise that a lot of environmental advocates are uncomfortable with. Luna: Speaking of something we value: the ability to explore these trade-offs without any ads or sponsors pulling the strings. We keep the podcast ad-free because we think that makes for better conversations. If you've gotten something out of episodes like this, you can support the show at buy me a coffee dot com slash fexingo. Lucas: And we genuinely mean it — every bit helps keep us independent. Now, back to the grid: one of the more fascinating data points I've seen is from the IEA, which recently raised its forecast for global data center electricity consumption. Luna: What did they say? Lucas: They projected that data centers could consume up to 1,000 terawatt-hours by 2026, which is roughly equivalent to the entire electricity demand of Japan. And AI workloads are the biggest driver of that growth. Luna: So Japan's worth of electricity, just for data centers. That's staggering. And it's not just the computing — it's the cooling. These facilities generate immense heat, and cooling them takes a lot of power too. Lucas: Right. And that's where geography comes in. Some of the biggest data center hubs are in places like Northern Virginia, where the grid is already strained. Dominion Energy, the local utility, has had to delay some interconnection requests because the transmission lines can't handle the load. Luna: So what's the fix? More transmission? More local generation? Both? Lucas: Both. And it's going to take regulatory changes too. Right now, getting a new transmission line approved can take 10 years. That's way too slow for the pace of AI investment. There's talk of federal permitting reform, but it's moving slowly. Luna: Meanwhile, data center developers are getting creative. Some are looking at on-site gas turbines, almost like mini power plants. Others are colocating with existing industrial facilities to use their excess steam or power. Lucas: And some are even considering small modular nuclear reactors, but again, that's a 2030s solution at best. For the next five years, it's going to be gas, gas, and more gas. Luna: Which ties back to the oil and gas stock rally we've been seeing. XOM up 2 percent in five days, CVX up 2.8 percent. It's not just about geopolitical risk — it's about real demand from the tech sector. Lucas: Exactly. And I think that's the key takeaway: the AI boom is not just a tech story. It's an energy story. And it's reshaping the economics of natural gas in a way that nobody predicted even three years ago. Luna: So as we look ahead to the rest of 2026, what are you watching? The price of Henry Hub? The interconnection queue? The utility stocks? Lucas: All of the above. But if I had to pick one, it would be the permitting reform debate. If the US can streamline the approval process for gas pipelines and transmission lines, that would unlock billions in investment. If not, we could see constraints that slow down AI development itself. Luna: And that's the ultimate irony: the technology that's supposed to accelerate everything could be held back by the physical limits of power infrastructure. Lucas: Exactly. AI might be virtual, but the electrons it needs are very real.