I normally chronicle the oilfield activity taking place in the shale basins of the U.S., often spending much of my time on companies operating in West Texas and New Mexico. A few counties in this area are the site of the Permian basin and the source of millions of barrels a day of oil and gas equivalent production thanks to fracking technology. In this article, I am going to sidestep that niche, but only for a while. If you read patiently to the end, I’ll swerve back into my normal lane and finish up in the oilpatch.
The low prices for gas have put of crimp in the fortunes of most of the drillers who focus on that commodity. I discussed this at length in a recent OilPrice article, which you should definitely read for more background on just what is ailing these companies. In this article, I am going to discuss a new source of demand for electricity that is causing power-generating companies to rethink what their true generating capabilities need to be in the near future. The answers they come up with could bode well for the beleaguered gas industry.
Artificial Intelligence-AI, sometimes referred to as Generative AI, or Gen-AI, meaning essentially the machine can think for itself- the truly profound change from the past, applications are exploding in scale and scope from a seeming virtual standstill a year or so ago. Along with them are requirements for new data centers to support this cloud-based activity. It is estimated that one new data center is being built every three days globally. The transformative promise of AI has the potential to bring about acceleration in human productivity equivalent to the arrival of the internet- now thirty years on. AI will touch every aspect of our lives. But it will need the power to do this. A lot of it.
“The surge in AI-driven power demand comes as other factors converge to create new strain on the grid. A wave of manufacturing plants are being developed across the U.S., spurred by new tax policies under the Inflation Reduction Act, and many states are working to use more electric power for transportation, heat and heavy industry.”
China Heavily Subsidized BYD to Expand Its EV Market Share
A recent WSJ article on this topic noted the demand from the AI sector isn’t happening in a vacuum. Power demand is on the rise in other sectors as well, and companies like Dominion Energy, (NYSE:D) are worried about their ability to deliver it, particularly in the face of net-zero 2050 commitments.
“We’re going to be net-zero by 2050. We still absolutely believe that,” said CEO Robert Blue. “But the demand growth now makes that more complicated.”
The AI power demand landscape
A recent Forbes article discussed the energy requirements of an AI data center and how they differ from a traditional data center. Legacy data centers include five to 10 kilowatts per rack as an average power density. AI applications require far more than 60 kilowatts or more per rack. To compound the problem, AI applications generate far more data than other types of workloads. This means that significant amounts of power-sucking data center capacity are required just to support them. All require a constant flow of electricity to keep running and for cooling. Just how much?
Barron’s in an article on AI power demand takes a stab at forecasting the near-term demand that AI will place on the grid. A source quoted in the article commented that:
“The overall power consumption increase will come on two fronts: an increase in the number of GPUs sold per year and a higher power draw from each GPU. Research firm 650 Group expects AI server shipments will rise from one million units last year to six million units in 2028. According to the article, most AI GPUs will draw 1,000 watts of electricity by 2026, up from the roughly 650 watts on average today.”
That is a range of 335 to 390 TeraWatt-TWH, hours of new demand by the end of this decade that was largely unanticipated by utilities just a few years ago. In a 2021 energy demand forecast, the EIA projected demand in 2030 of about 6 Trillion cubic feet of gas as a fuel source for about 30% of U.S electric power generation in 2030, of the 4.6 TWH we will need to power our lives. That has no allocation for newly recognized impending demand for AI. To the coming demand potential of 390 TWH for AI in 2030, how much increased demand for U.S. gas could we see?
Too much. If all 390 TWH were to come from natural gas another 36 BCF/D would be required. For reference according to the most recent EIA-914 the U.S. is producing about 125 BCF/D with YoY growth of about 6-BCF/D. When you add the 11 BCF/D coming from LNG exports that RBN Energy projects in the next couple of years to current needs for summer cooling demand, the ends of these two pieces of string do not meet in the middle. By a long shot.
Summing up
The entirety of this new demand will certainly not be wholly met with gas. We couldn’t do it if we wanted to. Renewables will play a role, but even combined, they will fall short of meeting AI 2030 requirements. The lack of electrical infrastructure for renewables generating transmission may lengthen timelines for new AI data center construction as well. Data center construction firm, CBRE Group notes on its site that timelines in some areas may be 2-6 years, citing approvals, cost inflation, and capital as the limiting factors.
“The approval process for wind, solar, storage and hydro-renewable projects now takes more than four years on average, with project costs more than doubling since 2020. In addition to delays, renewable energy developers are being asked to contribute capital for the buildout of high-voltage power lines and substations to get their projects on the grid.”
In that scenario, it’s no wonder that people are chatting up gas drillers for AI data center applications. Toby Rice, CEO of EQT Corp, (NYSE:EQT) was quoted at the CERAweek conference in Houston as saying-
“Tech firms building data centers are inquiring about buying gas from EQT. Rice said he got the same two questions at the conference: “How fast can you guys move? How much gas can we get?”
According to Rice, tech companies need reliable power, which renewable sources such as wind and solar can’t always provide because of the vagaries of weather. And large-scale nuclear facilities, only one of which is under construction in the U.S., have historically been expensive and time-consuming to build.
“Tech is not going to wait seven to 10 years to get this infrastructure built,” Rice said in an interview. “That leaves you with natural gas.”
Your takeaway
For investors a potential market tightening could spell opportunity in upstream gas driller’s stocks. As noted previously gas drillers are currently curtailing capex to cut the supply of gas in the coming months due to low prices. This fall 2-3 BCF/D could be coming from a new LNG plant start up. If we see summer demand for cooling in excess of what is forecast, we could begin to chip away at the storage volumes helping to depress prices.
EQT is trading at a 3X to EV/EBITDA. Most other gas drillers are in the same boat. If we do see a rally in gas prices driven by total demand, those multiples will expand as cash flow increases. Investors willing to make a bet on higher gas prices could be pleasantly surprised as the cash register starts to ring in companies like EQT.
By David Messler for Oilprice.com