On LNG, AI, and Shale Supply – We Believe the Turn in North American Natural Gas is Here

On LNG, AI, and Shale Supply – We Believe the Turn in North American Natural Gas is Here

Prefer to listen to this article? Please click the play button above.

The article below is an excerpt from our Q1 2024 commentary. 

We believe the North American natural gas market has reached a turning point. Significant shifts now taking place carry profound investment implications for the next twelve months. Our stance, which was cautious for almost a decade, turned bullish in the first quarter of 2020 when Henry Hub was at $1.43 per thousand cubic feet (mcf). In our 1Q20 letter, we wrote an essay titled “The Bull Market is Here,” stating: “If production were ever to falter, a massive bull market would result. That moment has arrived.” By summer 2022, Henry Hub gas surged more than six times, eventually reaching $9.68 per mcf as Russia’s invasion of Ukraine spread fears of a global natural gas shortage. However, back-to-back warm winters and an ill-timed fire at a US LNG export facility overwhelmed the nascent bull market and gas retracted its entire move, selling off to within five cents of the June 2020 low. Although it may seem like a strange time to write a bullish lead essay on natural gas, it was equally strange to make a bullish call in the spring of 2020, immediately before gas rallied 553%. 

Our research suggests this rally could be even stronger.

Bearish sentiment reached a fevered peak in the first quarter of 2024. The 2023-2024 North American winter was 5% milder than normal – on par with the extremely warm winter of 2011-2012. A lack of heating demand pushed prices sharply lower. Gas for delivery at Henry Hub fell 30%, broke $1.50 per mcf three times, and eventually bottomed at $1.48 per mcf on March 26th. Over the past twenty-five years, Henry Hub only broke $2.00 five times: in 2002, 2012, 2016, 2020, and earlier this year.

By some measures, natural gas pessimism in March was the worst in history. Gas bottomed amid a broader energy sell-off in 2002, 2016, and 2020. This year, natural gas reached $1.48 per mcf, with oil trading at a healthy $81 per barrel. Although a barrel of oil contains six times as much energy as an mcf of natural gas, WTI traded for an incredible fifty-two times Henry Hub natural gas – a nearly 90% discount on an energy-equivalent basis. Relative to oil, gas has only ever been cheaper once before, following the equally mild 2011-2012 winter.

Fig 1 - Oil-Gas Ratio and Henry Hub Discount to Energy Equivalent

Although the level of investor pessimism is comparable, today’s North American natural gas market could not be more different than in 2012. Although the 2011-2012 winter was also 5% milder than average, North American gas inventories ended the 2012 heating season 900 bcf above average, compared with a surplus of 500 bcf today. In 2012, the shale boom was on the verge of ushering in an unparalleled surge in US gas production. The combination of horizontal drilling and hydrological fracture stimulation allowed independent exploration and production companies to produce gas from previously impermeable source rock. 

After falling by 35% between 1970 and 2005, production first stabilized in the mid-2000s. In the eight years leading up to 2012, US dry gas supply grew by 14 bcf/d. Over the next eight years, it would increase by over 32 bcf/d, an incredible 50%. Throughout this period, production often grew by more than five bcf/d year-over-year. Several times, production growth exceeded 10 bcf/d in a single year. The incredible turnaround was entirely driven by the shales; conventional gas production continued to fall by 32% between 2012 and 2020. In total, shale production went from nothing more than a far-fetched dream to a 70 bcf/d behemoth by 2020. The gas shales produced more energy than Saudi Arabia’s prolific oil fields on an energy-equivalent basis.

Fig 2 - Average Annual Natural Gas Production Growth

The flood of shale gas overwhelmed the US market. Although utilities switched from coal to natural gas wherever possible, power generation could not absorb the excess. Crucially, the US could not export natural gas for most of this period; the first Lower-48 export terminals did not commence operations until 2017. With no outlet for the surplus production, North American natural gas prices decoupled from the rest of the world and they regularly began trading at a 40% to 60% discount on the world price.

According to the latest data, February production remained nearly 1 bcf/d lower than December – the sharpest non-weather-related slowdown outside of COVID-19 since 2008. Instead of lacking LNG infrastructure, the US is now the world’s largest gas exporter, with new terminal capacity set to surge over the next twelve months. More recently, analysts predict the promise of data center proliferation, driven by the rapid adoption of large language models, will usher in the most significant increase in domestic gas demand in US history.

The US is set to shift from a prolonged period of acute oversupply to a structural deficit of historic proportions. Although inventories remain high, our models predict they will draw to dangerous levels much sooner than anyone believes possible. Given this backdrop, it is unfathomable to us that US natural gas should trade at a record discount to its energy-equivalent price, even considering two consecutive mild winters. Investors should take note.

In many respects, the current natural gas market represents the perfect storm: dry gas production is faltering just as demand is set to surge. We have warned for several years that shale growth would slow. Our neural network models indicated that, although immense, the shale basins were not infinite. The Barnett and Fayetteville were the first two shale gas basins developed in the middle 2000s. Each field ramped up sharply before unexpectedly plateauing and declining by over 50%. We concluded that both fields peaked precisely once half their recoverable reserves had been produced, just as Hubbert’s theories predicted. By applying these same principles to the Marcellus, Haynesville, and Permian (collectively 75% of total shale gas production ) we warned growth would soon begin to slow before production rolled over entirely in 2025. It appears we were too conservative; US gas supply has likely peaked already.

As we mentioned, the US produced nearly 1 bcf/d less dry gas in February than the peak in December. Preliminary data suggests production declines will accelerate. The U.S. Energy Information Administration (EIA) expects June shale production to fall by another 2bcf/d compared with February.  Our internal models confirm the persistent declines. If the expectations are accurate (and we believe they are), total US dry gas production likely will fall by over 2 bcf/d between December 2023 and June 2024 – the sharpest six-month decline outside of COVID-19 since our data began.

Although it garnered no attention, in the latest Short-Term Energy Outlook (STEO), the EIA predicted that full-year 2024 dry gas production would fall by 1% compared with 2023. For this to happen, production must drop by an incredible 7 bcf/d between February and December 2024, breaking below 100 bcf/d for the first time since June 2022. The EIA expects US dry gas production to rebound sharply, making a new all-time high of 105 bcf/d in 2025, but we believe this prediction is too optimistic. The shales would have to arrest their declines and add nearly 1 bcf/d each month next year.

The EIA expects Henry Hub gas will average $3.10 per mcf in 2025, too low to elicit the huge drilling rebound needed to bring about this change. Between June 2021 and January 2023, gas averaged $5.67 per mcf, never once dropping below $3.10 and yet production growth only averaged 380 mmcf/d per month. Production grew at less than half the rate implied in the EIA’s 2025 projections despite a gas price that averaged 82% higher.

Fig 3 - Historical and Projected US Natural Gas Production

The only time monthly shale growth ever consistently approached 1 bcf/d was in 2017-2019; however, the shales were far less developed than today. In 2017, we estimate the Marcellus had only produced 30% of its total recoverable reserves, compared with over 50% today. Similarly, the Haynesville and Permian had made 25% and 30% of their recoverable reserves, respectively, compared with 50% today. Last decade, the Fayetteville and Barnett started their declines once half their reserves were produced. Our models suggest the same thing will happen with the Marcellus, Haynesville, and Permian. As a result, we believe the massive rebound the EIA expects next year will not be possible.

Making matters worse, the average well quality in the Marcellus, Haynesville, and Permian is steadily deteriorating, another indication of imminent field exhaustion. Cumulative six-month gas production per lateral foot in the Haynesville is 5% below the peak set in 1Q21. In the dry gas section of the Marcellus, productivity is 19% below the 4Q21 peak, while in the liquids-rich section of the play, productivity is 3% below the 2Q22 peak. In the Delaware side of the Permian, productivity peaked in 2019 and is currently 11% lower. In the Midland, productivity peaked in 2Q2020 and is currently 12% lower. Between 2017 and 2019, productivity steadily improved, providing a solid tailwind for monthly production growth. With productivity declining, we believe the same robust growth will be impossible.

Just as supply is set to falter, demand is expected to surge. The most critical driver is LNG terminal capacity. Over the next eighteen months, exports will increase by 4 bcf/d as three new domestic projects come online. Plaquemines is expected to commence commercial operations in the third quarter, ramping to 1.3 bcf/d. Corpus Christi will start next at 1.3 bcf/d, followed by Golden Pass in 2025 at 1.4 bcf/d. By mid-2027, an incremental 5 bcf/d of additional capacity is expected to come online, bringing total LNG exports to an incredible 20.4 bcf/d compared with less than 12 bcf/d today, the sharpest three-year growth in US history. Furthermore, both Canada and Mexico are sanctioning new LNG export capacity that could impact US supply. LNG Canada’s $30 bn Kitimat project will start up later this year, reaching its 1 bcf/d Phase I capacity in 2025. Although this gas will be sourced from Western Canadian Sedimentary Basin fields, it can potentially impact the nearly 8 bcf/d currently imported via pipeline from Canada. New Fortress Energy is almost ready to commission its Mexican Altamira project, which will liquefy 1bcf/d of US gas imported via pipeline. Along with the US terminals, these two projects will further tighten the North American market.

In addition to export demand, domestic gas consumption for electricity is expected to rise materially in the coming years, driven by the proliferation of data centers and artificial intelligence. Over the past several months, we have read countless articles detailing the energy demand from generative AI, such as ChatGPT. Some of the best work comes from Rob West at Thunder Said Energy who quantified the potential impact. Although he is uncertain about some of his projections, demand will be material. Modern artificial intelligence consists of two distinct phases: training and inference. During the training phase, vast quantities of computing power optimize trillions of parameters (or neurons) across zettabytes of textual data. This process consumes an enormous amount of energy. It is estimated that training GPT-4 alone consumed 50 GWH of electricity, equivalent to the average annual consumption of 5,000 American households. Once a model has been trained, end users queried it, a process known as “inference.” Although each inference requires only a fraction of the energy needed for training, a single model might be queried billions of times. 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.

Although it is beyond the scope of this essay to dissect each assumption, a few key drivers are worth discussing. First, many analysts expect energy efficiency to improve, mitigating energy demand growth. Unfortunately, this violates Jevons Paradox – a concept discussed in our 3Q23 letter. Jevons observed in the seventeenth century that instead of lowering demand, improved steam-engine efficiency dramatically increased English coal consumption. Although steam engines were becoming far more efficient, lower operating costs increased their proliferation, offsetting any gains and increasing overall coal demand.

Jevons Paradox is even more pronounced with generative AI. Supercomputer energy efficiency is measured in giga-flops per watt. Despite having improved five times since 2018, the total energy required to train an AI model has increased by an incredible 5,000 times. 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.

Second, despite the rhetoric around “green” data centers, we expect generative AI electricity demand will fall primarily to natural gas for two reasons. First, West estimates that the cost of training an AI model is five times more sensitive to electricity utilization than to price. As a result, both wind’s and solar’s inherent intermittency preclude them from being viable sources of electricity to power AI data centers. Second, even when a PV solar or wind installation generates power, the “quality” of the electricity, measured by its harmonic distortion, is unsuitable for the sensitive hardware used to train and query AI models. 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.

The impact of AI’s relentless power demand is already being felt. In May 2024, Dominion announced that new data centers in Virginia, used to train and query AI models, require several gigawatts of power, equivalent to several large nuclear power plants. t However, we do not expect either of these technologies to be rolled out until at least the end of the decade. In the interim, we believe natural gas demand will surge.

Although estimates vary, West believes 150 GW of AI data centers will be required by 2030, consuming 1000 TWH annually. Assuming 40% of global data center capacity is installed in the United States, AI data centers will consume 400 TWH of electricity, requiring 7 bcf/d of natural gas. Such a buildout would represent the largest increase in gas-fired power capacity in US history.

While West focuses on the energy needed to operate AI models, Mark Mills has tried to quantify the energy required to build the computer infrastructure. In his book, The Cloud Revolution, Mills outlines the massive energy needs of modern computing infrastructure. Although he notes that companies often do not fully disclose the energy needed to build modern data centers to power AI models, we have attempted to quantify the energy required to build out 150 GW of AI data center demand by 2030. Manufacturing high-performance semiconductors is energy-intensive. Although our numbers are preliminary, we believe manufacturing a modern data center consumes at least 8,500 MWh per MW of capacity. Reaching 150 GW of newly installed capacity by 2030 will consume an additional 2,500 TWH of electricity, or 430 TWH per year – nearly 50% as much energy as required to power the centers. Although most semiconductor fabs are located outside the United States, a strong push exists to relocate manufacturing domestically. Regardless of where the chip manufacturing is situated, it is clear that building out 150 GW of data centers will require a tremendous amount of energy, further tightening global energy markets.

In our view, US natural gas demand is set to grow at the fastest rate in history between now and 2030. At the same time, dry gas production appears to have peaked. While analysts are hopeful that a rebound is forthcoming, we are not as optimistic. The shale gas revolution resulted in a dramatic increase in supply, but as we have argued, immense is not the same as infinite. More than half of reserve estimates have now been produced in every major shale basin, an event that has corresponded historically with falling production. If our models are correct, and we believe they are, the most significant gas demand increase in history will occur just as production begins to falter.

US natural gas inventories are currently 2.7 tcf, nearly 600 bcf above seasonal averages. Although high by historical standards, our models suggest inventories will normalize as we progress through 2024 and into 2025. Assuming average weather, we expect inventories to work off all their surpluses early next year. By next fall, inventories could stand at record deficits. Although weather always remains a wild card in natural gas markets, new demand from LNG and AI combined with slowing shale production will likely overwhelm any period of mild weather.

Fig 4 - US Natural Gas Inventories

Today’s North American natural gas market bears no similarity to 2012. And yet, Henry Hub natural gas traded recently at a 75-90% discount to both the seaborne LNG and oil-equivalent price. Simply put, this is not sustainable. We believe natural gas-related equities represent exceptional value today. While gas stocks have dramatically outperformed the gas price since the start of 2023, their valuations remain incredibly depressed. For example, Range Resources today trades 30% below where it did in 2012 when natural gas prices were comparable. Over that period, proved reserves per share increased by 138%and the balance sheet improved. We calculate that the debt-adjusted SEC PV-10 per share increased nearly f ive- times since 2012. At current levels, we calculate Range Resources is pricing in a realized price of $2.62 per mcf, in line with the depressed spot price. If Henry Hub gas prices rebounded to only $4.00, Range’s debt-adjusted SEC-PV10 value would exceed $80 per share compared with $36. At $8.00 gas, in line with world prices, Range would be worth over $200 per share.

Although we have been very early, we believe North American natural gas, with less liquefaction and transportation, will converge with the global price, currently $10 per mcf. Investors are extremely bearish after two back-to-back mild winters but are neglecting the bullish shifts in both supply and demand currently underway. This is the most asymmetric investment we can recall. 

Intrigued? We invite you to download or revisit our entire Q1 2024 research letter, available below.   

2024.Q1 Research - On LNG, AI, and Shale Supply: We Expect the Turn in US Gas is Here.


Registration with the SEC should not be construed as an endorsement or an indicator of investment skill, acumen or experience. Investments in securities are not insured, protected or guaranteed and may result in loss of income and/or principal. Historical performance is not indicative of any specific investment or future results. Investment process, strategies, philosophies, portfolio composition and allocations, security selection criteria and other parameters are current as of the date indicated and are subject to change without prior notice. This communication is distributed for informational purposes, and it is not to be construed as an offer, solicitation, recommendation, or endorsement of any particular security, products, or services. Nothing in this communication is intended to be or should be construed as individualized investment advice. All content is of a general nature and solely for educational, informational and illustrative purposes. This communication may include opinions and forward-looking statements. All statements other than statements of historical fact are opinions and/or forward-looking statements (including words such as “believe,” “estimate,” “anticipate,” “may,” “will,” “should,” and “expect”). Although we believe that the beliefs and expectations reflected in such forward-looking statements are reasonable, we can give no assurance that such beliefs and expectations will prove to be correct. Various factors could cause actual results or performance to differ materially from those discussed in such forward-looking statements. All expressions of opinion are subject to change. You are cautioned not to place undue reliance on these forward-looking statements. Any dated information is published as of its date only. Dated and forward-looking statements speak only as of the date on which they are made. We undertake no obligation to update publicly or revise any dated or forward-looking statements. Any references to outside data, opinions or content are listed for informational purposes only and have not been independently verified for accuracy by the Adviser. Third-party views, opinions or forecasts do not necessarily reflect those of the Adviser or its employees. Unless stated otherwise, any mention of specific securities or investments is for illustrative purposes only. Adviser’s clients may or may not hold the securities discussed in their portfolios. Adviser makes no representations that any of the securities discussed have been or will be profitable. Indices are not available for direct investment. Their performance does not reflect the expenses associated with the management of an actual portfolio.