The article below is an excerpt from our Q3 2025 commentary. 

We have long taken a certain perverse pride in conducting research that is not merely original, but occasionally so unfashionable that polite company feels compelled to edge toward the exits. It has led us, more than once, to conclusions that bear little resemblance to the consensus view. Our aim, unfailingly, is to peer just far enough into the future to spot emerging of trends not yet acknowledged—ideally early enough to produce superior investment returns.

This work, needless to say, is neither leisurely nor without hazard. Many market observers prefer the pleasantries of the crowd, where reputational safety lies in numbers. Others make a vocation of narrating yesterday’s events and then insisting—with admirable confidence— that they had foretold the entire affair. And then there is the dominant strain on Wall Street: extending today’s trends into the indefinite future, as though history were not littered with the corpses of straight-line assumptions.

Our approach attempts—sometimes awkwardly—to stand apart. We recognize that any honest attempt to predict the future must occasionally collide with error. When that happens, as it inevitably will, we try to identify our missteps quickly and adjust. Keynes, with characteristic dryness, put it best: “When the facts change, I change my mind. What do you do, sir?”

We invoke that sentiment now in response to the Substack commentator known as Doomberg, who recently took issue with our expectation that U.S. shale gas production would soon stop growing and eventually roll over.

In January 2024, we debated Doomberg on Adam Taggart’s Thoughtful Money podcast. During that discussion, we made what was—at the time—a highly unpopular assertion: that both shale oil and shale gas production were nearing their respective peaks. Shale oil was then growing at an impressive 1 million barrels per day, more than half of it from the Permian Basin. We emphasized, almost to the point of repetition, that the Permian was nearing its apex. Few agreed. Yet by October, the basin did the unthinkable—it peaked. Twenty months later, the EIA reports Permian crude output down 100,000 barrels per day year-on-year and total shale oil down 160,000 barrels per day.

In that same interview, we predicted that gas production would follow oil downward, though on a slight delay. That call has aged less gracefully. As we will explain shortly, the error arose from a misunderstanding of the Permian’s associated gas behavior—specifically, the way its wells tend to grow gassier as they mature. Outside the Permian, shale gas did exactly what we expected. It rolled over soon after our interview aired and remains down 1.35 bcf/d. Although production has staged a partial rebound from earlier lows, shale gas output beyond the Permian has not reached new highs.

We remain accountable for every call—those that proved correct and those that did not. For that reason, we make all of our old quarterly letters publicly available, allowing anyone to trace the evolution of our thinking. Doomberg has recently done just that, prompting us to revisit several earlier projections.

Two regions deserve special attention: the Marcellus and Permian associated gas—each a dominant contributor to U.S. natural gas supply growth over the past decade.

In our 1Q20 letter, we estimated that the Marcellus would ultimately recover 92 trillion cubic feet of gas. That forecast emerged from a deep neural network built in late 2019 and refined through early 2020—trained to assess ultimate recovery by analyzing where a well was drilled, how it was drilled, and how it was completed. From there, we estimated remaining drilling locations and their expected productive capacity.

Drawing heavily from King Hubbert’s teachings on depletion, we observed that hydrocarbon basins often peak once half their recoverable reserves have been produced. Based on that framework, we expected the Marcellus—then producing 23.5 bcf/d and averaging more than 2 bcf/d of annual growth—to peak and roll over in 2021.

History, in its usual wry fashion, had other ideas. We were wrong about the timing. And yet something unmistakably changed in 2021. Productivity, measured as initial production per lateral foot, crested and began to decline. Output continued to rise briefly, eventually reaching a monthly high of 27.8 bcf/d in 2023. But the torrid growth of the preceding decade flatlined. For the past two years, Marcellus production has barely budged.

In 1Q22, we revised our ultimate recovery estimate higher—by 40%, from 92 to 132 tcf. By then, we had significantly enhanced our modeling. We incorporated detailed geological data that had previously been unavailable and altered the framework in a subtle but profoundly important way. Instead of modeling total well output with lateral length as an input, we normalized production per lateral foot. That simple reframing—removing length from the list of things the model had to “learn”—freed it to capture the real drivers of productivity: geology and completion techniques.

The result was a material improvement in accuracy and a sharp upward revision in estimated recoverable reserves. The facts had changed. So did our minds.

Our revised reserve estimates neatly clarified why Marcellus production managed to keep climbing between 2020 and 2022. Yet with the new figure—132 tcf rather than the 92 tcf we had worked with earlier—we arrived at much the same destination. By 2022, it appeared that within a year roughly half of the basin’s recoverable gas would be produced, and growth would cease. This time, the model proved truer to life. In 2023, right on cue, the Marcellus reached its monthly production peak.

The next turn in our thinking came later in 2023. We had begun reexamining how we model undrilled locations—long the most intricate and stubborn piece of the puzzle. Mapping every theoretical wellbore, especially in heavily developed acreage, proved both computationally exhausting and increasingly unsatisfying. So we shifted course. Instead of hunting for individual future wells, we measured the total remaining lateral footage in a given area. Subtract what has already been drilled, apply a representative production profile for that specific neighborhood, and the estimate of remaining recoverable reserves becomes far more tractable. The change nudged our reserve figures higher once again, though not nearly to the extent of earlier revisions.

Our most substantial revision arrived in late 2024, when we retired the deep neural network in favor of a boosted forest algorithm. Accuracy improved once again, but the real triumph was interpretability: we could now see, with far greater clarity, what forces had shaped productivity per lateral foot over time. At the same moment, we revisited how the model treated the later years of a well’s life. Shale wells burst onto the scene with extraordinary rates, only to decline sharply—often delivering more than 80% of their ultimate reserves in the first twenty-four months. Because that early flush dominates the economics, we had historically devoted most of the model’s attention to it, leaving the out-years somewhat neglected. But then we noticed an unexpected development—the tails were lengthening.

This realization carried two notable consequences. The obvious one was yet another upward revision in recoverable reserves. The subtler, and ultimately more important, was its effect on our understanding of when a field actually peaks. If each well settles into a longer, flatter tail of modest production, total recovery increases, but the timing of the peak does not shift much—because the basin’s growth is still governed by those explosive early months. In practical terms, a field may now reach its high-water mark after producing only 35% of its total recoverable reserves, rather than the 50% we once assumed. The peak arrives on schedule, but with far more gas still technically in the ground.

Taking all of these refinements together, we now estimate that the Marcellus will ultimately yield roughly 210 tcf—well above the 135–150 tcf range we projected in 2023. Yet the implication for timing remains largely unchanged. Expecting the field to peak after 50% of 150 tcf has been produced is functionally identical to expecting a peak after 35% of 210 tcf has been recovered. The model is more accurate, the reserves more generous, but the broad production profile of the basin remains the same.

We have never hesitated to refine our models, nor should any analyst who hopes to remain grounded in reality. But perpetual improvement does not negate the facts on the ground. Something unmistakable happened in the Marcellus in 2023—per-foot productivity declined, and the basin’s once unbroken ascent came to a halt. Whatever the refinements in methodology, the underlying explanation, in our view, is straightforward: depletion has arrived.

The Permian Basin is, in many respects, the more intricate—and more captivating—story. Doomberg notes that in 2020 we suggested Permian gas production would soon roll over, and he is correct that, rather than declining, it has since doubled. But the remark deserves its proper setting. In the depths of the COVID-induced oil price collapse, we argued that drilling activity in the oilier reaches of the Permian would inevitably slow, and that associated gas would decline alongside it. We were equally clear that the basin itself had not yet reached its summit. Our expectation then, as now, was that Permian oil—and therefore gas—would likely peak in 2025 or 2026.

Curiously, while Permian oil production did indeed soften during COVID, gas output never paused—it kept rising. The same paradox confronts us today. Oil production peaked twelve months ago, yet gas growth has proceeded undeterred. In hindsight, this revealed a blind spot in our earlier thinking. Our models treated a well’s oil stream and gas stream as distinct forecasts. The reservoir, however, does not. The Permian is a true mixed-medium system, with substantial volumes of gas dissolved in the crude—one continuous hydrocarbon cocktail. Underground, pressure keeps the gas in solution, much like carbon dioxide trapped in a sealed bottle of soda. Only once the well is opened does the separation begin.

Once the well is drilled—or, to extend the metaphor, once the bottle is opened—pressure falls and the gas begins to break away from the liquid. Early in a Permian well’s life, when downhole pressure remains high, oil and gas travel upward together and only separate at the surface, where conditions are gentler. But as production continues and reservoir pressure declines, the balance shifts. Gas peels off first and flows preferentially up the wellbore. The older the well becomes, the gassier it gets.

Thus, while our neural network did a commendable job estimating each well’s ultimate oil and gas recovery, it overlooked a crucial dynamic: as the field ages—not as theory, but as thousands of individual wells—the production mix shifts inexorably toward gas. We had modeled the endpoints correctly, but not the journey.

A typical Permian well may begin life producing roughly 75% oil on an energy-equivalent basis. As it matures, that share can slip toward 45% before leveling off. Which means that as the production-weighted average age of the field rises, the gas-cut rises with it. During periods of rapid drilling, the average age can hold steady—or even decline—masking the shift. But once growth slows and declines emerge, the field inevitably gets gassier. This is precisely what has happened. Over the past several years, the Permian has managed to expand gas output at three times the rate of oil, adding another 1 bcf/d in just the last twelve months—essentially matching its long-term pace—even as oil production has begun to fall outright.

We have taken to calling this episode the great “gas burp.” Fortunately, the transition from oil to gas is anything but mysterious. Using a traditional model built on differential equations, we estimated both the duration and magnitude of post-peak gas growth in a field like the Permian. Our current view is that gas output could still rise by as much as 1 bcf/d over the next 12 to 18 months, before declining in tandem with oil. Given the underlying geology, it is exceedingly difficult to imagine Permian gas continuing to grow far into the future— despite what many analysts still contend.

Making bold forecasts can, at times, test one’s patience, but it helps to keep the broader picture in view. Yes, we have adjusted our models and lifted certain estimates, but taken as a whole, the neural network—and now the Boosted Forest—has served us remarkably well. When we first warned of a looming slowdown in late 2019, shale oil output was increasing by an average of 650,000 barrels per day each year from 2008 through 2019. Since then, annual growth has slipped to less than 200,000 barrels per day between 2019 and 2024, before turning negative—down 160,000 barrels per day over the past twelve months. Shale gas outside the Permian tells a similar story: average yearly gains of 4.6 bcf/d from 2008 to 2019 have dwindled to 1.4 bcf/d between 2019 and 2024, followed by a 1 bcf/d decline over the last eighteen months. We freely acknowledge our miscall on Permian gas, but we also believe we now understand the mechanism—and that its growth will slow sharply before rolling over entirely within the next year or two.

Crucially, whatever incremental growth remains in the Permian is likely to be canceled out by declines elsewhere. Other shale basins are already weakening, and their retreat will probably offset most—if not all—of the Permian’s gains, leaving total U.S. shale gas production flat at best, and more likely headed lower.

We make no claim to infallibility. But we do believe our work has offered a valuable compass for navigating U.S. energy markets. In many respects, shale gas today resembles shale oil when we debated the issue in early 2024: production was still rising briskly, yet subtle, easily overlooked signals suggested the tide was turning. Within a year, shale oil reached its peak and began to decline. We were proven right then, and we believe we are likely to be proven right again—this time with shale gas.

Turning to balances, as ever, the weather will have the final say as North America settles into the heart of heating season. Inventories now sit 138 bcf above the five-year average—a striking reversal from February, when they were 230 bcf below it. A mild March followed by an even milder summer swelled storage to nearly 200 bcf above average by September, only for an early November cold snap to spur demand and pull gas back out again.

Historically, between November 1 and April 1, storage falls by roughly 1.9 tcf—about 12.6 bcf per day. Should this winter merely mirror last year’s—which, it bears repeating, was still 1% milder than normal—withdrawals could easily reach 16 bcf per day, pushing inventories to nearly 300 bcf below average by spring. Against that backdrop, dry gas production is expected to average 108 bcf/d through April, while net exports should average 17 bcf/d, up from 13 bcf/d last winter as new LNG facilities ramp. Remarkably, LNG exports alone appear to have increased by 2 bcf/d over just the past two months. Last winter, consumption averaged 106 bcf/d under slightly milder-than-normal conditions; with incremental data center demand, a repeat could nudge that figure to 106.5 bcf/d. Add it all together, and inventories could draw by as much as 2.3 tcf this season—tightening relative storage levels by roughly 400 bcf.

Looking beyond winter, the arithmetic becomes increasingly uncomfortable. Rising LNG exports, growing data center demand, and largely stagnant shale supply make it hard to imagine inventories not beginning a steady drawdown relative to long-term norms. Weather will still have its say, of course, but the setup is already taking shape. And so one must ask: how much longer can U.S. natural gas trade at a 60% discount to global prices? The gas bull market may finally be upon us, though one suspects most investors will recognize it only after the price screens have already done the shouting.

Curious to learn more now?  Read more in our Q3 2025 research newsletter, available for download below.

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.