Why Natural Gas Will Be AI’s Next Great Shortage

Summary

• The core call: US natural gas breaks below all historical storage by 2029. After 18 months modeling every producing well and pipeline “at the atomic level,” Smith finds 2026-27 appropriately supplied — but plug in contracted LNG and AI compute and by mid-2028 storage “break[s] very materially below… where gas available in storage has ever been before.” Price upside is “unbounded and convex,” felt most acutely in electricity bills 2028-2030.
• The die was cast before AI. LNG exports go from 15 BCF/d today (on ~110-112 BCF/d production) to 35 BCF/d by end-2030 — sited, permitted, project-financed. Maximum US deliverability is 128-132 BCF/d, so the ~20 BCF/d of possible supply growth was already spoken for: “even without AI compute, we had sources and uses matched.” Compute’s ~5 BCF/d base-case pull (12-15 BCF/d unmitigated at looser probability screens) breaks the balance.
• Nobody is positioned. The curve is flat at $3.50-3.60 out to the 2030s, EKT (likely EQT) is shutting in gas, the rig count shows “no one’s on it,” and 2028 is illiquid. Once utilities roll forward, “you’re going to start to see a knife fight to secure natural gas physical in 28 like we really haven’t seen before” — the analog is DRAM: “slowly at first and then all at once,” and this market is “not two years out, but six plus months out.”
• Top longs: Expand at 4x EBITDA, Range, and a solar windfall. Expand Energy probably controls ~70% of remaining core Haynesville, is CEO-less after turnover, and trades ~4x forward EBITDA with a low-to-mid-teens FCF yield “on a gas forward curve where no one believes what I’m telling you”; Range is probably the highest-quality Appalachian upstream. Since gas sets marginal power price, free-fuel solar gets margin expansion for zero capex — XPLR (XIFR) marks PPAs to much higher values, and residential solar “grows exponentially from here.”
• The most viable long-term fix is large-scale nuclear online 2033-34 — AP1000s, not SMRs, whose companies Smith says are not set up to manufacture and truly scale to tens of gigawatts. Czar-for-a-day move: the US government builds 2-4 AP1000s end-to-end with the ~$260B Loan Programs Office to derisk the supply chain. Levered names: Kamico (likely Cameco; 49% of Westinghouse, “deeply undervalued” inside it) and BWXT.
• Losers: the US consumer first, then the turbine boom. Caterpillar — which he thinks is doubling Solar Turbines capacity by end-2029 — is “just at the exact wrong time”; Bloom Energy at 2 GW+ — which Smith doesn’t think can win gas in competition — won’t win gas in competition, and fuel cells get treated as backup gen only. Hyperscalers budget energy at ~10% of cost — if gas doubles or triples structurally, it’s 20-30% of compute cost by 2029, “at a time when they’re supposed to be reaching escape velocity with profitability.”

Deep dive

1. A historic deficit opens in 2028

• Smith’s conclusion after 18 months of well-level reconstruction: natural gas — over 40% of US power generation, “overtaking petroleum” as the country’s most important fuel — is fine in 2026-27, but assign compute to specific assets, keep exporting contracted LNG, and by mid-2028 storage “break[s] very materially below in a historical way where gas available in storage has ever been before.” By 2029 “we drop below all known historical storage evidence”; by 2030, ceteris paribus, storage “looks very, very low.”
• Every price precedent he can point to was transitory — Russia-Ukraine took gas to $8 or $10/MCF, the 2014 polar vortex and December 2022 hit $6, $8, or $10 — but “what we’re talking about are structural drivers of demand against a known possible production of gas, and they don’t match up.” He explicitly won’t put a price target on it: the deficit is “convex and unbounded,” and the transmission channel is electricity prices 2028-2030.

2. The die was cast before AI showed up

• The setup: shale abundance from 2010 turned the US from importer to exporter, from likely Cheniere’s early exports to 15 BCF/d of nameplate LNG capacity today — 12-15% of the ~110-112 BCF/d the US produces — with sited, permitted, project-financed buildout to 35 BCF/d by end of 2030. “The die has been mostly cast.”
• Against that, having evaluated every producing well (acreage digitized into polygons, well performance parameters known), Smith caps maximum US deliverability at 128-132 BCF/d — roughly +20 BCF/d, and “we will take the under on that.” LNG growth alone consumes it: AI compute is the marginal straw, not the whole load.
• Patrick’s naive fix — “just shut off the exports, who cares?” — gets a layered no: contract law, tens of billions of project financing, and the US becoming about a third of global gas supply that FTA and non-FTA allies depend on. “It could be stopped, but it’d be very complicated.” At $8-10, spot cargoes stop lifting — but “spot cargoes alone can’t solve this”; you’d have to break contracted cargoes to “some Japanese utility counterparty who has counted on it.”

3. Compute: 5 BCF/d credible, multiples of that proposed

• His base case takes only P50 projects — some approvals, usually a PPA, an interconnection agreement or one in process — and gets ~5 BCF/d of credible incremental demand from AI compute. Drop the screen to P30/P0 and it “more than double[s]” to 12-15 BCF/d by the early 2030s if unmitigated.
• Every proposed solution consumes more gas. Bloom’s 6-series fuel cell takes 150 million cubic feet a day per gigawatt, and the market prices ~2 GW/year of manufacturing ramping toward 5 — “there isn’t the gas for that unless you take it from something else.” Bring-your-own-generation, what hyperscalers are now asked to do, means more gas, not less: “every time you read a press release from Bloom or from GE Vernova (likely), just think more gas.”

4. Not a rock problem — a steel problem

• The resource is well mapped and he expects no rescue: companies burn through “most of the existing captured inventory in the next four or five years,” other known basins are uneconomic and unpiped, and “I do not think we’re likely to be surprised by some new major shale find.” His pointed diligence tip — some producers claim more well inventory “than we can justify with facts”; ask for exact engineered locations on a map.
• Above ground, three binding constraints: processing (2-3 years to build, insufficient even for his 20 BCF growth case — “build it now, imminently”), gathering (needs material investment immediately), and interstate pipelines, where the US has built essentially one in 10-12 years (Mountain Valley, Appalachia to Mid-Atlantic). “You don’t press a button to solve this.”

5. The market is asleep — the knife fight starts when 2028 rolls into view

• The tape shows total complacency: gas at $3.50-3.60 with a flat curve to the 2030s, EKT (likely EQT) shutting in production “because they think it’ll be more valuable later,” a rig count that says “no one’s on it.” Fifteen years of abundance means “gas has lulled everybody to sleep,” and “that complacency is going to take us right up to the point where it’s too late.”
• The 2028 strip is illiquid — that’s why the curve hasn’t moved. Once utilities turn the page and generators start securing supply, “you’re going to start to see a knife fight to secure natural gas physical in 28 like we really haven’t seen before” — and in six months, as companies roll forward, “you could start to see the forward curves really move up materially.”
• The analog came from one of Patrick’s recent guests, who heard the pitch and said “this sounds like DRAM two years ago” — “slowly at first and then all at once.” Smith’s sharpest line: “Imagine being short memory a year ago… That is what this natural gas market looks like to us, not two years out, but six plus months out.”

6. Winners: Expand at 4x EBITDA, Range, and a solar windfall

• Expand Energy is “far and away the biggest winner”: it probably controls ~70% of remaining core Haynesville — the key swing basin — currently CEO-less amid a search, stock down hard over six months, trading ~4x forward EBITDA and a low-to-mid-teens free cash flow yield “on a forward curve where no one believes what I’m telling you to be the case. The assets have not changed.” Range is probably “the highest quality upstream company in Appalachia” with real room to grow production and returns.
• The non-obvious leg: “as natural gas goes, power prices go in the country” — gas sets the marginal fuel in each power market, so free-fuel solar captures margin expansion for zero incremental capital. XPLR (ticker XIFR, formerly the NextEra yield co) marks its PPAs to market at much higher values without capex; Clearway (likely — heard as “Split away energy”) sits in a similar spot.
• Residential solar — one of the only ways to protect yourself against 10am-6pm peak pricing — “grows exponentially from here,” economic even after the first removal of install incentives since the late ’70s, especially paired with batteries.

7. The only real fix: large-scale nuclear by 2033-34

• There is “no silver bullet” and no bridge fuel other than solar and wind, so the answer is AP1000-class reactors online 2033 or 2034, “as soon as it can come on” — the deficit “only gets worse in 31, 32 and beyond.” SMRs get no quarter: “many of the SMRs are still science experiments,” not set up to manufacture the tens of gigawatts required.
• He doesn’t dodge the track record — two reactors in 30 years (Vogtle 3&4: ~3x cost, ~15 years birth to service), VC Summer nearly bankrupting SCANA — but counters with China building ~39 reactors, 34 of them 1 GW+, a third modeled on the AP1000: “we know a lot more today about building large-scale nukes than we did when these mistakes were made.”
• Czar-for-a-day: the US government should build 2-4 AP1000s entirely, beginning to end, using the Loan Programs Office’s ~$260 billion — 10-20 reactors are envisioned “but no one wants to be first.” Levered names: Kamico (likely Cameco) (49% of Westinghouse alongside Brookfield’s 51%, Westinghouse “deeply undervalued within Cameco”) and BWXT, primary nuclear supplier to the US Navy with “lots of dollar content in the AP1000s.”

8. Losers: the consumer first, then the turbine boom

• “Sadly… the biggest losers of this would be the US consumer” — an “awful trade-off” between exporting gas, powering AI compute, and household bills that will feed the NIMBYism already surfacing. “We’re not anti-AI… the US consumer is probably going to pay the bill in the meantime.”
• The gas-turbine and genset manufacturers are rerunning the early-2000s boom-bust: Caterpillar — which he thinks is doubling Solar Turbines capacity by end of 2029 — lands “just at the exact wrong time,” Bloom at 2 GW+ — which Smith doesn’t think can win gas in competition — won’t win gas in competition, and fuel cells get treated as backup gen only. Orders for even large-scale gas plants “could slow very meaningfully as 26 progresses,” and an E&C firm at 25x cash flow whose main business is building gas plants should focus on accretive M&A to backfill and diversify.
• For hyperscalers, energy is budgeted at ~10% of cost; if gas “double[s] or triple[s] structurally, even without weather,” it’s 20-30% of the cost of compute by 2029 (elsewhere he stretches to 40%) — while every LCOE decision today is underwritten on a flat forward curve.

9. The pushbacks he’s pre-modeled — and the homework for CEOs

• Permian rescue? Already in the model: the 7+ BCF/d of pipelines building 2026-2030 is included, and “if there were a new gas pipeline that would come on between now and 2030, we would know about it” from regulatory timelines. More Permian gas requires much higher oil for longer (oil sat at $55 before the Iran conflict) — “which only exacerbates this consumer crisis.” Batteries? Lithium-ion deployments are modeled; a step-function chemistry like sodium would be “a watershed moment that I would welcome.”
• Canada has “by far the deepest and richest resource of economic gas in the ground, but it’s been trapped behind pipe,” with storage about a fourth of the US’s — he’d build a 1-2 BCF/d line into the Midwest/MISO and wheel it around the power markets. Globally, failing to deliver the 30-35 BCF/d leaves Europe choosing between “Russian gas or much, much higher cost US gas” and potentially hurts important allies “at a time where we really want them to be allies.”
• His closing challenge to any CEO with energy as an input: know exactly where your physical gas comes from, lock supply, and understand counterparty risk two to three years out. “What if gas is not 350 but 10 or more?… Do you have a backup plan?” And a metric to watch: “performance per watt is probably a compute metric that we’re going to care more and more about.”