All podcasts / Invest Like the Best / Summary

Matthew Smith - Natural Gas: The Next Bottleneck

2026-07-21 - 58 min - source - Read full transcript

Patrick O'Shaughnessy (host)Matthew Smith

Key insights

The U.S. is on track to exhaust its working natural gas storage cushion by 2030, making the upside risk to gas prices unbounded and convex.

Chronometer's atomic-level modeling of every producing well, pipeline, and processing asset shows natural gas is adequately supplied through 2026-27, but starting in 2028 demand from AI data centers plus committed LNG export growth begins drawing down the 4 TCF U.S. working gas storage buffer faster than it can be replenished, breaking below all historical storage lows by 2029 and approaching critical levels by 2030.

natural-gas-supply-crunch

The setup for this crunch predates AI entirely - it was locked in by LNG export commitments made over the last decade.

U.S. LNG export capacity is set to roughly double from about 15 BCF/day today to 35 BCF/day by 2030 based on projects already sited, permitted, and financed years in advance. Combined with the roughly 20 BCF/day of new production the country can realistically bring online, exports alone already consume most of the available supply growth before any AI compute demand is added.

lng-exports

AI data center demand for gas is real but smaller in the base case than popularly assumed - the risk is in the tail, not the base case.

Chronometer's P50 (contracted, credible) estimate for incremental AI-driven gas demand is only about 5 BCF/day. But if you include every announced-but-unsecured project (a P30/P10 case), that figure could more than double to 12-15 BCF/day by the early 2030s - and there simply isn't spare gas in the system to cover the tail scenario without displacing something else.

ai-power-demand

Shutting off LNG exports to solve the domestic shortage sounds simple but is legally and geopolitically constrained.

Tens of billions of dollars in project financing and long-term contracts underpin U.S. LNG projects, and the U.S. now supplies roughly a third of global gas trade, making allies in Europe and Asia dependent on it. Curtailing exports would violate contract law and free-trade agreements and would be one of the last levers pulled, not the first.

lng-exports

The real bottleneck isn't gas in the ground - it's midstream infrastructure (processing, gathering, and interstate pipelines).

Chronometer's mapping shows plenty of resource remains in Appalachia, the Haynesville, and the Permian, and the U.S. isn't at risk of physically running out of gas over the next 20 years. The constraint is that processing capacity takes 2-3 years to build, gathering systems need material near-term investment, and only one new interstate gas pipeline (Mountain Valley) has been built in the last 10-12 years due to permitting barriers.

natural-gas-supply-crunch

Large-scale nuclear (AP-1000 reactors), not SMRs, is the only long-term solution Smith sees as viable at the scale needed.

Most SMR companies remain unproven science experiments not set up to manufacture at the tens-of-gigawatts scale required. Large-scale AP-1000 units, despite the troubled Vogtle 3 and 4 build (3x over budget, ~15 years), are better understood today and could realistically come online by 2033-2034 if construction starts now - which Smith argues is already too late to avoid a mid-decade gap.

nuclear-buildout

The U.S. consumer is the biggest loser of this dynamic, caught in a trade-off between exports, AI compute, and household electricity bills.

As gas tightens, the country faces an explicit trade-off between exporting gas to allies, powering AI data centers, and keeping consumer electricity prices in check. Most government-proposed fixes (like Bring Your Own Generator policies for data centers) actually increase gas consumption rather than reduce it, worsening the underlying deficit.

ai-power-demand

Utility-scale and residential solar are underappreciated winners because rising electricity prices increase their margins at zero incremental capex.

Companies like XPLR Infrastructure and Clearway Energy mark power purchase agreements to market, so rising prices flow straight to margin since their fuel (sunlight) is free. Residential solar, paired with batteries, is framed as one of the only ways individual households can hedge against the 10 a.m.-to-6 p.m. peak power price spikes Smith expects once gas gets tight.

energy-investing-winners-losers

Distributed gas generation - fuel cells and gas turbines built for fast time-to-power - are being overbuilt into a supply-constrained future.

Companies like Bloom Energy and Caterpillar (Solar Turbines) are expanding distributed gas-fired generation capacity aimed at hyperscalers, but Smith is skeptical there will be enough gas to run these assets reliably at scale; he expects orders for large-scale gas generation to slow meaningfully as the constraint becomes apparent through 2026-2029.

energy-investing-winners-losers

The gas-shortage dynamic is compared directly to the DRAM memory shortage - underinvestment now, sudden scarcity later.

Smith describes floating this thesis to industry contacts, one of whom likened it to the DRAM shortage: 'slowly at first and then all at once.' He estimates natural gas could rise from about 10% to 20-30% of hyperscalers' total cost of compute by 2029 if his structural price thesis plays out, directly compressing their economics.

ai-power-demand

The market hasn't started pricing this risk because 2028 gas contracts remain illiquid and nobody has started hedging yet.

Smith notes surprise that despite tens of billions of dollars flowing into distributed power assets that will need gas, almost no counterparties have moved to lock in physical gas supply contracts for 2028 and beyond. He expects a 'knife fight' to secure physical natural gas once utilities and generators start hedging in earnest, which should be the trigger that finally moves the flat forward curve.

natural-gas-supply-crunch

The clearest hedge against being wrong - a Permian oil/gas production surge - requires sustained high oil prices that themselves would worsen the consumer cost problem.

The main pushback Smith has heard is that Permian associated gas could fix the shortfall. His base case already includes ~7+ BCF/day of pipelines under construction through 2030, and any further Permian gas growth would require much higher and more sustained oil prices to justify additional drilling - which itself would compound the affordability problem rather than solve it.

natural-gas-supply-crunch

Media referenced

SemiAnalysis - other - Cited as the leading source for behind-the-meter power/load forecasts for AI data centers, but Smith notes it stops short of modeling where the natural gas to fuel that power will actually come from - the gap his own work fills.

Companies

Chronometer Partners - Matthew Smith's investment firm, which spent 16+ months modeling nearly every U.S. natural gas well, pipeline, and processing asset at the atomic level to build this thesis.
Positive Sum - Patrick O'Shaughnessy's firm; standard disclosure that clients may hold positions in securities discussed.
Expand Energy - Called the single biggest winner among gas producers - controls ~70% of remaining core Haynesville acreage, trading at 4x EBITDA despite a depressed forward curve; currently searching for a new CEO after leadership turnover.
Range Resources - Named as having the highest-quality upstream rock in Appalachia with significant room to grow production and returns.
EQT - Cited as evidence of producer complacency - currently shutting in natural gas because it believes it will be more valuable later rather than investing to grow supply now.
XPLR Infrastructure (XIFR, formerly NextEra Energy Yield Co) - Flagged as a solar winner - marks its power purchase agreements to market and captures a windfall as electricity prices rise, with no incremental capex.
Clearway Energy - Another utility-scale solar asset owner positioned to benefit the same way as XPLR from rising electricity prices.
Caterpillar (Solar Turbines) - Flagged as a likely loser - doubling distributed gas-turbine manufacturing capacity right as gas may become too scarce/expensive to fuel those assets.
Bloom Energy - Flagged as poorly positioned - its fuel cells need gas that Smith argues won't be available at scale; also exposed to rare-earth supply risk in manufacturing.
Cameco - Owns 49% of Westinghouse (with Brookfield owning 51%); called undervalued given the coming large-scale nuclear buildout Smith expects.
Brookfield - Co-owner (51%) of Westinghouse alongside Cameco, positioned to benefit from renewed large-scale nuclear construction.
BWXT - Primary nuclear supplier to the U.S. Navy; called a beneficiary of the coming nuclear cycle given its content in AP-1000 reactor builds.
Westinghouse - Maker of the AP-1000 reactor design Smith sees as the only credible large-scale solution to the coming power deficit.

Techniques and frameworks

P50/P30 probability-weighted demand modeling - Chronometer's method for sizing incremental AI-driven gas demand: only counting projects with signed PPAs and interconnection agreements in the P50 base case (~5 BCF/day), versus a P30 extreme case (12-15 BCF/day) that includes announced-but-unsecured projects.
Atomic-level well and asset mapping - Digitizing every company's acreage as lat-long polygons against measured well-performance data to independently estimate maximum U.S. gas deliverability (128-132 BCF/day), rather than trusting company-reported inventory claims.
Levelized cost of energy (LCOE) - Standard framework hyperscalers use to evaluate power sources; Smith argues it currently looks favorable for gas only because everyone is pricing off a forward curve that assumes gas stays cheap and flat through the 2030s.

Summary

Matthew Smith, founder and CIO of Chronometer Partners, returns to Invest Like the Best six years after his last appearance to lay out an 18-month research effort concluding that the United States is heading into a historic natural gas deficit by 2028-2030, with electricity prices as the primary transmission mechanism. His team built a bottom-up model of nearly every producing gas well, pipeline, and processing asset in the country, digitizing company acreage as mapped polygons against measured well-performance data rather than trusting company-reported inventory claims. The conclusion: the U.S. can add roughly 20 BCF/day of new gas production, but committed LNG export growth (from ~15 to 35 BCF/day by 2030) alone consumes most of that headroom before any AI-driven demand is added.

Critically, Smith argues the setup for this crunch predates AI - it was locked in by a decade of LNG export buildout following the shale boom that began around 2010. AI data center demand is real but modest in his contracted base case (about 5 BCF/day), though it could balloon to 12-15 BCF/day in a tail scenario that includes all announced-but-unsecured projects. The deeper structural constraint isn't gas in the ground - Smith is confident the resource exists - but midstream bottlenecks: processing capacity that takes years to build, underinvested gathering systems, and an interstate pipeline network that has added just one major line (Mountain Valley) in over a decade due to permitting barriers. As working gas storage draws down starting in 2028, breaking below all historical lows by 2029, Smith expects prices to become "unbounded and convex," pointing to $8-10+/MCF precedents from the Russia-Ukraine gas shock and past weather anomalies as reference points, though he's cautious about setting a hard price target.

The conversation maps winners and losers across the value chain. Upstream, Expand Energy (control of core Haynesville acreage) and Range Resources (Appalachia's highest-quality rock) stand out as producer winners trading at depressed multiples the market hasn't repriced. On the power side, solar - both utility-scale players like XPLR Infrastructure and Clearway Energy, and residential solar paired with batteries - benefits from rising electricity prices without needing new capital, since their fuel is free. Nuclear is framed as the only durable long-term fix: Smith dismisses most SMR companies as unproven science experiments and instead backs large-scale AP-1000 reactors (Westinghouse, majority-owned by Brookfield and Cameco, plus supply-chain beneficiary BWXT), arguing construction needs to start now to have new capacity online by 2033-2034. On the loser side, he flags distributed gas generation manufacturers like Bloom Energy and Caterpillar's Solar Turbines business, which are scaling fast-time-to-power assets into a future where he doubts there will be enough gas to run them reliably, plus the U.S. consumer generally, who absorbs the cost as the country balances exports, AI compute, and household bills.

Smith closes by noting the market hasn't begun pricing this risk - 2028 gas contracts remain illiquid, and despite tens of billions of dollars committed to gas-dependent power assets, almost nobody has moved to lock in physical supply. He likens the dynamic directly to the recent DRAM memory shortage: underinvestment compounding quietly until it breaks "all at once." His practical advice to any CEO with energy as a cost input: understand exactly where your physical gas will come from, assess counterparty risk over a 6-month-plus (not multi-year) horizon, and stress-test plans against gas prices of $10/MCF or more rather than today's $3.50.

Notable Quotes

"We are headed into a place where we see an historic deficit in natural gas supply available in the United States, which does portend some pretty serious consequences." - Matthew Smith

"The die was cast long before AI compute came to the scene." - Matthew Smith

"This sounds like DRAM two years ago, slowly at first and then all at once." - Matthew Smith, recounting a colleague's reaction to his thesis

"Imagine being short memory a year or 18 months ago and finding out all of a sudden you're short memory. That is what this natural gas market looks like to us, not two years out, but six-plus months out." - Matthew Smith

"Sadly, the biggest loser of this would be the U.S. consumer." - Matthew Smith