I have three children, the oldest of which is finishing up her junior year of high school. We are early enough in the college process where anything is possible. AP exams are being finished, SATs have been taken, application essays are being brainstormed. This time next year, the reality of low acceptance rates will set in and there may be some disappointment. 

Similarly, when it comes to the potential impact of artificial intelligence (AI), the world is just beginning to consider what is possible. We don’t yet know how much power AI will need 10 years from now and how real the demand is for more natural gas to support data centers tied to that technology at some point in the future.

With earnings this quarter, utilities and midstream companies have started rolling out estimates and frameworks for considering the scale and total addressable market (TAM) potential of that new demand source. 

Load growth

Demand for electricity in the U.S. has been stagnant for a long time, to the tune of roughly 0% growth in the last decade and just 0.5% annually for the last 20 years.

Several new trends have emerged that appear to be changing that trajectory, including:

  • Growth in industrial/manufacturing production from reshoring supply chains;
  • Electrification of transportation and other things; and
  • Data center power demand from the explosion of generative AI applications. 

Load growth is here now, with power demand growth expected to finish above 2% in 2024.  There is potential for much higher load growth by the end of the decade, maybe as much as 5% annually, driven by 15% CAGR in data center power demand. Such a steep change in demand for electricity has implications across the energy value chain that will require massive investment in power generation, around $50 billion estimated by Goldman Sachs. 

That level of power generation investment, assuming 60% is natural gas-fired, would lead to around 3.3 Bcf/d of incremental natural gas needed, according to Goldman Sachs. Citi has run numbers on AI demand, coupled with slower retirement of existing coal plants, and a series of other assumptions that pegs a base case of incremental natural gas demand from AI data center demand by 2030 of 3.8 Bcf/d in a base case or more than 5 Bcf/d in an upside case with greater than expected data center growth.

Recently, the market seems to have had an awakening as to how much infrastructure would be needed to supply that much power, and utilities are suddenly the best performing sector in the market after being largely ignored for 18+ months. 

Electricity demand is growing three times faster this decade than in the prior decade. Natural gas as a percentage of total U.S. power generation is at an all-time high already at more than 40%.

That electricity demand needs reliable natural gas to function as coal capacity is retired and less reliable renewables take a greater share of the power supply stack.

Things that could throw cold water on this demand growth party:

  • Ability for utilities to permit and build out new power plants, challenges when it comes to rolling that demand into rate base;
  • Natural gas supply challenges, related to permitting new pipelines; and
  • Technology company budgets and/or demand, or maybe even pushback on carbon emissions.

Technology companies that have made carbon reduction goals may need to walk back energy transition goals which seem at odds with a ramp-up in natural gas-based power. That dynamic may be why technology companies have been so aggressive in contracting for new renewables related to data center demand.

What are midstream companies saying?

On this quarter’s earnings calls, midstream companies were ready to discuss AI data center demand growth. It was popular for sell-side research analysts to tally up the number of times data center was mentioned on calls (using AI to do so in some cases). According to Wolfe Energy’s equity research team, “data center” was mentioned on first-quarter conference calls 86 times overall, including 26 in prepared remarks and 60 times in Q&A. 

As far as how much opportunity there is, Kinder Morgan Chairman Rich Kinder offered up his answer to the AI demand question like a recent MBA interviewing for a job at an investment bank: “AI demand alone is projected at about 15% of demand in 2030. If just 40% of that AI demand is served by natural gas, that would result in an incremental demand of 7 [Bcf] to 10 Bcf a day.”

So, the demand potential is somewhere between around 3.5 Bcf/d and 10 Bcf/d by 2030, with most folks coalescing around 5 Bcf/d to 6 Bcf/d. Either end of that range would be big if it comes true, but I will take the under if we’re talking about by 2030, even though I might believe the high-end number is realistic if we add another 10 years. 

In the near term, the upstream and midstream sections of the natural gas value chain are happy to have some hope for future demand growth to hold up the longer end of the natural gas futures curve, keeping the curve in contango in a year when Waha natural gas prices have at times been negative and Henry Hub has struggled to break $2/MMBtu. Rather than rely exclusively on global demand for LNG for incremental demand, adding in the potential for a meaningful domestic demand source is welcome.