Smarter Than You - AI, Data Center Power Demand and the Implications for Natural Gas

Data center power demand is soaring as AI — artificial intelligence — rapidly expands across all sorts of applications. That statement is certainly the top buzz factor in today’s energy markets. These facilities need 24x7, super-reliable power, and there’s only one power generation fuel that has any hope of keeping up with the demand surge: natural gas. While most data center developers would prefer green energy to cover their power requirements, the intermittent nature of wind and solar means that for many facilities, it can't happen, at least for the short-to-medium term hyped-up market we are seeing right now. But how much incremental power are we talking about? And how much natural gas will be needed? That’s what we’ll explore in today’s RBN blog.

“Ever wondered why your smart speaker keeps asking for more power? It’s not just craving a caffeine boost — it’s trying to keep up with the insatiable demand of AI’s appetite for electricity!” Those aren’t our words. Instead, they’re what ChatGPT gave us (exclamation point included!) when we asked it to write a humorous first sentence for an RBN blog about AI and power demand. Not bad, really, but we’ll stick with human bloggers. (Well, for the time being at least!)

Seriously, it’s a feeding frenzy out there — AI is the next big thing. It’s just like Wall Street’s obsession with Nvidia. It seems like energy markets are salivating over ChatGPT and anything related to generative AI, with energy BizDev reps looking to reel in some of those high-demand, energy-hog projects. That’s definitely the sense we get from RBN’s informal crowd-sourced market intelligence system — meaning questions and comments from our 45,000 blog recipients. Mostly, you all want to separate the hype from the reality and figure out what’s really going on in this market.

In response, we’ve been doing our own serious research to help put AI energy demand in perspective. Although we pulled information from many sources, much of what we’ve included in this blog is sourced from a very useful white paper published by EPRI — the Electric Power Research Institute — in May 2024 titled “Powering Intelligence — Analyzing Artificial Intelligence and Data Center Energy Consumption.” We took up the challenge to validate, triangulate and corroborate the paper, and to make the information a bit more digestible for natural gas folks who might consider providing fuel/infrastructure for power plants needed to meet AI demand.

U.S. Data Center Market

As of March 2024, the U.S. had 5,381 data centers, nearly doubling from around 2,700 in January 2021. This substantial growth in data center capacity has been driven largely by the increasing demand for what are generally classified as AI-powered tasks, such as speech recognition, image recognition, predictive analytics, personalized diagnostics/treatment, logistics/mapping applications, fraud detection, and of course, generative AI. The rapid rise in interest in generative AI is particularly noteworthy, catalyzed by the sudden success of ChatGPT (which launched in November 2022) and a few other AI chatbots that have popped up riding the ChatGPT wave, including Copilot, Claude, perplexity.ai and Jasper.

While the national-level growth estimates are substantial, the geographic concentration of the industry and the local challenges it creates are even more striking. As shown in Figure 1, nearly 50% of 2023 U.S. data center load was concentrated in only four states: Virginia, Texas, California and Illinois. Including another 11 states brings the total to 80% of the load. Virginia alone accounted for 22% of the national data center demand, benefiting from strong internet connections, minimal disruptive events, an available skilled labor force, proximity to population centers and users, and reliable backup power sources (for more, see Storm Front).