In a recent OilPrice article, I argued that new demand from AI Data centers would be bullish for the natural gas market. In this article, I will extend that discussion and explain why I believe there is a dramatic reset coming for gas-focused E&P stocks. Exactly when isn't part of the discussion here, but I will lay out a compelling investment thesis for stocks exposed to this commodity.
It's tough to be a gas bull, but it's natural gas, I mean. Over most of the past decade, that point of view has pretty well guaranteed you a broken heart and a wounded wallet down the road. It's not always very far down the road, depending on where you made your move in the cycle. That's how it's always been, but I think the bull case is strengthening despite the prodigious inventory overhangs that are now controlling the price narrative. I hesitate to use language like "it's different this time." But what are the ramifications if it is?
First, a little history. Seasonal temperatures have historically driven much of the pricing dynamic with gas. Everybody knows that when it got cold, gas prices went up as people heated their homes to stay warm. At the end of heating season, prices dropped as the thermostat switched to air conditioning with rising temperatures. King coal took over for the summer and fall.
In recent years the LNG exports have created new gas demand, which was fortunate as the peaking of the shale revolution has brought new supplies of "associated gas" into the market. This is gas that, until recently was flared, if existing gathering and trunkline infrastructure wasn't available. Flaring is bad on a lot of levels, and thousands of miles of midstream infrastructure have been built and are continuing to be built, ensuring this precious resource is utilized as efficiently as possible.
Power generation has quietly begun to fill a gap left between coal plant retirements and renewable generation's inability to meet summer cooling demand. Last year, a record number of coal plants went offline, and gas demand rose by several percent to fill the gap. In fact, Bloomberg reported in an article recently that new demand from AI Data Centers has forced utilities to delay coal generating plant closures.
Demand from AI Data centers is so new on the scene that it hasn't factored in utility company demand forecasting to this point. That's about to change as energy providers like Dominion Energy reassess their needs to meet this new source of demand, but the situation may be more complex than and energy-intensive than has previously been widely discussed. Jevon's paradox, first applied to coal consumption in steam engines, also applies to data consumption. Brian Janous, founder of Cloverleaf Infrastructure commented on this principle in a recent Bloomberg article.
"Jevons said in regard to the notion that efficiency increases in steam engines would result in the use of less coal. 'No, that's not what's going to happen. We're going to use more coal because we're going to mechanize more things.' And that's exactly what we do with data just because we've had Moore's Law for years, and so chips has become incredibly more efficient than they were decades ago, but we didn't use less energy. We used much more energy because we could put chips in everything."
About a year ago, I felt that with cracks emerging in the renewable energy narrative, that gas would find new demand from new sources. It hasn't happened with the rapidity I expected, as gas prices are still in the dumper, but when you dig into the details, the outcome appears inevitable. In that scenario, many of the gas-oriented energy equities we follow are drastically undervalued at present levels. A recent note from natural resources analyst firm Goerhring & Rosencwajg (G&R) probed this idea in some detail and cited work by Rob West of Thunder Said Energy.
Essentially, AI power demand comes in two phases: Training and Inference. The Training part consumes massive amounts of data on the front end as AI models scrape the internet, absorbing data. West calculates that "training Chat-GPT-4 alone consumed 50 GWH of electricity, equivalent to the average annual consumption of 5,000 American households."
Then comes Inference as the models are queried, each query consuming a tiny amount of energy but multiplied by billions of queries for Chat-GPT-4 alone.
"West estimates a ChatGPT "inference" requires ten times as much energy as a Google search --3.6 Wh compared to 0.3 Wh. Generative AI's total energy consumption is a function of several related variables: the number of new models trained per year, the complexity of each model, the energy efficiency of new AI chipsets, and the total queries per trained model."
Wrapping up
A few key takeaways come from this research. First as the G&R note states, most analysts are not taking Jevon's into account when making their forecasts and are assuming energy efficiency will improve and mitigate energy demand growth. A direct contradiction of Jevon's. It gets worse.
"Training GPT-4 required fifty times more energy than a 2022-vintage model. As chips become more energy efficient, model complexity grows exponentially, requiring more energy to train. Furthermore, the number of distinct models has also grown exponentially. A significant number of more complex models has dwarfed any improvement in chipset energy efficiency, a trend that we expect will continue."
West's work suggests that the cost of training an AI model is 5X more sensitive to electricity utilization than to price. The intermittency of wind and solar generation makes them not viable sources for this purpose. In fact these sources by their nature introduce what West refers to as "harmonic distortion," adding to their unsuitability for the sensitive hardware used in training AI models. G&R concludes this section of their Q-1 note with a bullish forecast for gas.
"As a result, we believe the widespread proliferation of AI must be met with either coal, natural gas, or nuclear-based power. It is unlikely that new coal-fired power will be sanctioned in the US and the lead time on new nuclear power plants is too long to meet demand over the next several years. Therefore, natural gas should be the primary beneficiary of the AI rollout through the decade's end."
Finally, this burgeoning new demand is going to be met with stagnating the best case, or declining availability of new gas, as this most recent edition of the EIA-914 reveals.
Now you should see why we are bullish on gas and think so many of the companies we follow represent outstanding entry points if this thesis is correct. Take pure-play Marcellus producer EQT, (NYSE: EQT) as an example. At just $4.50 per MCF for gas on the NYMEX contract, EQT should trade for $53.00 per share with just 10% of their risked locations factored in. You can run this calculation with any gas-focused E&P that strikes your fancy.
I see the coming AI gas demand as an asymmetric opportunity for investment at the current discounted levels that many gas-focused E&Ps carry. In that scenario, the broken hearts of gas bulls will be mended.
By David Messler for Oilprice.com
More Top Reads From Oilprice.com:
Mr. Messler is an oilfield veteran, recently retired from a major service company. During his thirty-eight year career he worked on six-continents in field and… More