Recent OilPrice article, I have argued that new demand from AI data centers is bullish for the natural gas market. In this article, I expand on that argument and explain why I believe we are on the verge of a dramatic reset in gas-focused E&P stocks. While I won’t discuss the exact timing here, I do present a compelling investment thesis for stocks tied to this commodity.
It’s tough to be a gas bull, but it’s natural gas after all. For much of the past decade, that view has almost guaranteed heartbreak and bruised wallets down the road, not necessarily in the not-too-distant future, depending on where you acted in the cycle. As has been the case in the past, I think bulls are strengthening their position despite the massive inventory glut that is currently driving prices. I hesitate to use phrases like “this time is different.” But if it is, what are the implications?
First, a little history. Historically, seasonal temperatures have been a big factor in gas price trends. We all know that when it gets colder, gas prices rise as people heat their homes to stay warm. At the end of the heating season, prices fall as thermostats switch to air conditioning as temperatures rise. In the summer and fall, coal was king.
This has been fortunate, as LNG exports have created new gas demand in recent years and the peak of the shale revolution has brought new supply.Associated Gas“It’s been brought to market. This is gas that until recently was flared where existing gathering and trunking infrastructure was not available. Flaring is bad in many ways, and thousands of miles of midstream infrastructure have been built, and continue to be built, to ensure this precious resource is utilized as efficiently as possible.
Electricity generation is quietly filling the gap created by coal plant closures and renewable energy generation to meet summer cooling demand. A record number of coal plants were closed last year, and gas demand increased by several percent to fill the gap. In fact, Bloomberg reported in a recent article: New demands from AI data Centers forced power companies to delay the closure of coal-fired power plants.
Demand from AI data centers is still new and has not previously been factored into utility demand forecasts, but that is changing as energy providers adopt AI data centers. Dominion Energy reassess needs Meeting this new source of demand is likely to be more complex and energy intensive than has been widely discussed so far. Jevons The paradox that first applied to coal consumption for steam engines also applies to data consumption. Cloverleaf Infrastructure This principle has been recently Bloomberg article.
“Jevons said about the idea that more efficient steam engines would reduce coal use: ‘No, that’s not going to happen. As mechanization increases, coal use will increase.’ And that’s exactly what we’re doing with data. Because Moore’s Law has been going on for so many years, chips are incredibly more efficient than they were a few decades ago, but they don’t use less energy. Because we can put chips into everything, they use a lot more energy.”
About a year ago, I Cracks appearing in the renewable energy narrativeGas will find new demand from new sources. Gas prices remain sluggish and haven’t materialized as quickly as we expected, but digging into the details, the outcome seems inevitable. In that scenario, many of the gas-related energy stocks we follow are significantly undervalued at current levels. Recent Notes Rob West of natural resource analysis firm Goerhring & Rosencwajg (G&R) has explored this idea in detail: Sandor said energy.
Essentially, AI’s electricity demand is split into two phases: training and inference. The training part consumes a lot of data on the front end as the AI model scrapes the internet to absorb data. West calculates that “Chat-GPT-4 training alone consumed 50 GWH of electricity, equivalent to the average annual consumption of 5,000 US households.”
Inference is then performed when the model is queried: each query consumes a tiny amount of energy, which is amplified to billions of queries for Chat-GPT-4 alone.
“West It is estimated that ChatGPT’s “inference” requires 10 times more energy than a Google search (3.6 Wh versus 0.3 Wh). Generative AI’s total energy consumption is a function of several relevant variables, including 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.”
summary
Several key takeaways emerge from this research: First, as the G&R memo notes, most analysts don’t take Jevon’s predictions into account when making their forecasts, and instead assume that improvements in energy efficiency will moderate the growth in energy demand, which is in direct contradiction to Jevon’s predictions.
“Training GPT-4 required 50 times more energy than the 2022 model. As chips become more energy efficient, model complexity grows exponentially, requiring even more energy to train.” Additionally, the number of different models has increased exponentially. More complex models are outpacing improvements in chipset energy efficiency, and this trend is expected to continue.”
West’s research shows that the cost of training AI models is five times more sensitive to electricity availability than price. Wind and solar power are intermittent, making them unsuitable for this purpose. In fact, these sources tend to produce what West calls “harmonic distortion,” making them unsuitable for the sensitive hardware used to train AI models. G&R concludes this section of its first quarter report with a bullish prediction for gas:
“As a result, we believe widespread adoption of AI will need to be met by either coal, natural gas, or nuclear power. It is unlikely that any new coal-fired power will be approved in the U.S., and the lead times for new nuclear plants are too long to meet demand over the next few years. Natural gas should therefore be a major beneficiary of AI adoption. The end of the decade.”
Finally, this surge in new demand will, at best, stagnate or result in reduced new gas supplies. EIA-914 reveal.
If this theory is correct, then you can see why we are bullish on gas and why we believe many of the companies we follow are excellent entry points. Take pure-play Marcellus producer EQT (NYSE: EQT) as an example. NYMEX contracted gas is just $4.50 per MCF, but if you factor in 10% of risk location, EQT should be trading at $53.00 per share. You can do this calculation with any gas-focused E&P of your choice.
I believe future AI gas demand will present an asymmetric investment opportunity at the discount levels that many gas-focused E&Ps are currently trading at. In that scenario, gas bulls’ broken hearts will be licked.
Article by David Messler of Oilprice.com
Other popular articles from Oilprice.com:
