Everything About AI
NextFin News - Energy Transfer LP has formally committed $1.6 billion to build a 215-mile natural gas pipeline across the Southeast, the latest and clearest signal yet that the AI buildout is being financed in steel and pipe, not just in chips. The South Mississippi Pipeline, approved by the Dallas-based midstream partnership on Tuesday, will move up to 1.2 billion cubic feet of gas per day from eastern Louisiana across Mississippi, with Energy Transfer owning 51% and its customers funding the rest. The decision lands as data centers and liquefied natural gas export terminals reshape the U.S. gas market faster than most investors priced in a year ago.
The question this project answers — and raises — is whether the AI-driven demand for natural gas is a durable structural shift or a cyclical wave that will crest once the current data center rush slows. Energy Transfer's answer is unambiguous: it is betting billions that the demand is real, long-dated, and contracted. The market, for its part, has rewarded that posture. Energy Transfer units have climbed roughly 34% over the past year, and the partnership reports more than $2.7 billion in additional data-center-linked pipeline projects already under construction in Texas and Arizona.
The Deal: What Energy Transfer Actually Committed To
The South Mississippi Pipeline is a 215-mile (346-kilometer) interstate line that will connect the Carthage Hub and the Haynesville production area — two of North America's most prolific gas supply basins — to markets across the Southeast. Its designed capacity of 1.2 billion cubic feet per day (Bcf/d) is not a speculative bet on spot volumes: Energy Transfer said it will operate the project and retain a 51% stake, with the remaining equity coming from customers who have signed up for the capacity. That customer-funded structure is the financial architecture of the entire AI-gas thesis — the shippers, not the pipeline builder's unitholders, carry the volume risk.
The project follows an open season that Energy Transfer ran for the line earlier this year, when it described the route as providing firm transportation from the Carthage and Perryville hubs to multiple Southeast markets. At that stage the company said capacity could start at 1 Bcf/d but was scalable to more than 2 Bcf/d, depending on market response. Tuesday's commitment locks in the larger configuration.
For Energy Transfer, South Mississippi is one spoke in a much wider wheel. The partnership has roughly 140,000 miles of pipeline and associated energy infrastructure across 44 states, with more than 55 Bcf/d of interstate and intrastate throughput capacity and about 236 billion cubic feet of storage. Its 2026 growth program is dominated by projects aimed squarely at power and data-center demand: the $2.7 billion Hugh Brinson Pipeline and up to $5.6 billion for the Desert Southwest expansion, both designed to feed gas-fired generation in Texas and Arizona.
The Demand Engine: Why Gas, and Why Now
The pipeline's economics rest on a single, testable premise: AI data centers will consume natural gas at a scale that reorders the U.S. supply-demand balance for the rest of the decade. The numbers behind that premise have been revised sharply higher.
RBC Capital Markets, in a May 2026 strategy note, forecast natural gas consumption from data centers reaching approximately 6.1 Bcf/d by 2030, with a likely range of 6 to 7 Bcf/d by the end of the decade — roughly a 20% increase to baseline power demand from 2025 levels. Even that, the bank noted, remains secondary to the liquefied natural gas export capacity additions expected through 2030. Beyond 2030, RBC expects zero-carbon energy sources to gain relative share, but sees natural gas retaining a significant operational role given current technology and infrastructure timelines.
A research report released this year went further. It projected that gas consumed for electricity generation at U.S. data centers will rise by 15 Bcf/d in the decade ending in 2035 — a volume that, even after discounting for projects that may never be completed, exceeds the current gas consumption of every country except China, Russia, Iran, and the United States itself, based on U.S. Energy Information Administration data. Natural gas is expected to supply 69% of the power needed by new grid-connected facilities, the report said, because of its abundance, low production cost, and the ability of gas-fired plants to ramp output quickly to match the 24/7 load profile of hyperscale data centers.
The combined effect is a demand shock with two engines. The same report projects power-sector gas consumption reaching 54 Bcf/d by 2035, up 18 Bcf/d from 2025, while gas demand from LNG exports rises by 21 Bcf/d over the same period. U.S. producers are expected to increase output by 35 Bcf/d between 2025 and 2035 — leaving an 11 Bcf/d gap between projected supply growth and the combined AI-plus-LNG demand. Henry Eaton, the gas market analyst who led the research, cautioned that the data-center forecast carries "fairly large" error bars in both directions. That caveat matters: the entire midstream buildout is being sized against a demand curve that its authors openly describe as uncertain.
Against that backdrop, U.S. marketed natural gas production averaged 116.8 Bcf/d in the second quarter of 2025, and the EIA expects output to remain near that level through 2026, averaging around 116 Bcf/d in both years. In other words, the supply side is essentially flat while the demand side is being rewritten.
The Competitive Race: Midstream's $5.1 Billion Pivot
Energy Transfer is not alone. Williams Companies has committed more than $5.1 billion to what it calls a "power innovation" business model, centered on building modular, gas-fired power plants directly at data center sites. Its flagship Project Socrates — a $1.6 billion investment set for completion in the second half of 2026 — will provide committed power generation and associated gas pipeline infrastructure for a specific, unnamed investment-grade client. The two Socrates sites in New Albany, Ohio, will deliver a combined 400 megawatts of behind-the-meter power to a Meta-affiliated data center campus under a 10-year fixed-price power purchase agreement, with an option to extend. Williams has described the project's roughly 5x EBITDA build multiple as implying a return of about 20%.
That is the strategic shift in one sentence: midstream companies are no longer just moving molecules. They are selling reliability, speed-to-market, and integrated power — and charging for it. For Williams, Project Socrates was its first deal of this kind. For Energy Transfer, the pivot began earlier, with a string of direct data-center supply agreements that read like a customer roll call of the AI buildout:
- CloudBurst Data Centers (February 2025): Energy Transfer's Oasis Pipeline will provide up to 450,000 MMBtu per day of firm natural gas supply to a Next-Gen Data Center campus outside San Marcos, Texas — enough to generate up to approximately 1.2 gigawatts of direct, behind-the-meter power for at least 10 years. The company called it its first commercial arrangement to supply natural gas directly to a data center.
- Fermi America (October 2025): a 10-year agreement to provide an initial supply of approximately 300,000 MMBtu per day to Fermi's HyperGrid AI campus outside Amarillo, Texas, described as the world's largest behind-the-meter AI private grid campus.
- Oracle (November 2025): natural gas supply to three U.S. data center sites, two in Texas, totaling approximately 900,000 Mcf per day.
- Entergy Louisiana (November 2025): a 20-year binding agreement for 250,000 MMBtu per day of firm transportation service via the Tiger Pipeline to fuel Entergy's power plants in North Louisiana, which support Meta's hyperscale AI data center in Richland Parish.
"We couldn't be more upbeat about where we sit today with data centers throughout the country, but especially in Texas, because of our ability to perform and provide reliable gas service. We're very excited for the future and we expect data centers and power plants to be part of our growth for many years."
Mackie McCrea, Energy Transfer's co-chief executive officer, said that on the company's third-quarter 2025 earnings call. Co-CEO Tom Long added:
"These opportunities continue to show how extensive our interstate pipeline network is throughout the country and how fortunate we are to have so many of them near our pipeline assets."
The Second-Order Effect: Pipelines Become the Bottleneck
The first-order read of this news is straightforward: more data centers mean more gas demand, which means more pipeline construction. The second-order effect is more consequential, and less widely appreciated. The constraint on AI's energy appetite is no longer just generation capacity — it is transportation capacity.
A data center can finance a gas-fired turbine in months. A pipeline that crosses state lines, secures firm shipper commitments, and navigates the Federal Energy Regulatory Commission's certificate process takes years. That timing mismatch gives the owners of permitted, contracted interstate pipe a form of leverage that pure power generators do not have: once a data center campus is anchored to a specific pipeline route, switching costs become prohibitive, and the pipeline's contracted capacity becomes a scarce, toll-collecting asset. This is the mechanism by which a midstream partnership converts a technology boom into decades of fee-based cash flow — and why Energy Transfer is willing to let customers fund nearly half the equity of a $1.6 billion project rather than rush to build it alone.
There is also a geographic arbitrage at work. The Southeast and Texas, where most of the near-term data center load is clustering, sit on top of the Haynesville, Permian, and Appalachian supply basins. Moving gas from those basins to the load centers is not a matter of building new long-haul trunk lines from scratch; it is a matter of adding lateral capacity — exactly what South Mississippi does. The projects with the shortest path to revenue are the ones that connect existing supply hubs to existing demand pockets, which is precisely the profile of Energy Transfer's sanctioned backlog.
The Counter-Thesis: What Could Break the Trade
The strongest argument against this thesis is not that AI demand is imaginary — it is that it is already priced in, and that the demand curve is far less certain than the capital being committed to it. Environmental groups have already begun challenging the wave of Southeast pipeline authorizations. The Sierra Club, reacting to the Federal Energy Regulatory Commission's approval of two Southeastern pipelines in August 2026, called them "dangerous and unnecessary" — a signal that permitting risk, while currently favorable, is not zero.
More importantly, the demand forecasts themselves carry wide error bars. The lead author of the data-center research explicitly flagged "fairly large" uncertainty on both the upside and downside. If AI model efficiency improves faster than deployment — if the compute-per-watt curve keeps falling — the 6 to 15 Bcf/d of incremental data-center demand could arrive later, or smaller, than today's sanctioned pipe assumes. And if the buildout does slow, midstream pipelines face a specific kind of pain: their costs are largely fixed, and a line built for 1.2 Bcf/d that runs at half capacity does not earn half its return — it can earn far less, because the debt service and contracted capacity obligations do not scale down with utilization.
The counter-thesis also has a valuation dimension. Energy Transfer units trade at roughly 14 times trailing earnings with a dividend yield near 6.7%, and the stock is up about 34% over the past year. A substantial portion of the "AI gas" narrative is already reflected in that price. The margin of safety for new capital depends on the contracted, fee-based nature of the cash flows holding up — not on spot gas prices or on the uncontracted portion of future demand.
Cyclical or Structural? The Verdict
So which is it — a cycle or a regime shift? The answer is both, and the distinction determines where the risk sits.
The structural leg is real and durable. Data centers are a new, large, 24/7 load class located in gas-rich regions, and the infrastructure that serves them — interstate pipelines with multi-decade lives and contracted capacity — cannot be replicated quickly. LNG export capacity, the other engine of demand, is already permitted and under construction; those terminals will run for decades once built. The rules of the system — Federal Energy Regulatory Commission certification, state siting, long-term contracts — favor incumbents with permitted routes and anchor shippers. That is a regime change in the U.S. gas market, and it will not revert on its own.
The cyclical leg is the pace and timing of the AI buildout itself. History offers a cautionary pattern: the U.S. shale boom of the 2010s produced a wave of sanctioned midstream capacity that ran ahead of realized demand, leaving some assets underutilized until catch-up growth arrived. The current data-center rush could follow a similar S-curve — a burst of announcements, a period of project cancellations and delays as power constraints bite, then a slower, more selective phase of completion. The 2026–2028 window is the most crowded; the 2030-plus window is far less certain.
The right way to underwrite Energy Transfer's decision, then, is to separate the two. The contracted, customer-funded portion of South Mississippi and the existing data-center supply agreements are structural — they are locked in by long-term contracts and will produce fee-based cash flow regardless of the AI hype cycle. The uncontracted, expansion-capacity portion is cyclical — it depends on the next wave of data center announcements actually reaching final investment decisions. Energy Transfer's 51% stake with customer-funded remainder is the market's way of expressing exactly that split.
What to Watch Next
Three signals will determine whether this thesis is playing out as expected or rolling over:
- Contracted capacity on South Mississippi. If Energy Transfer reports that the 1.2 Bcf/d line is fully subscribed at or before its in-service date, the structural thesis is confirmed. If subscription lags, the cyclical risk is materializing.
- AI data-center power demand versus the 6–15 Bcf/d forecasts. A downward revision from RBC Capital Markets or the U.S. Energy Information Administration — specifically, data-center gas demand tracking below 4 Bcf/d by 2028 — would be the clearest falsifying signal for the structural-demand call.
- Permitting outcomes at FERC and in the states. Continued approvals of Southeast and Texas pipelines support the buildout timeline; a string of delays or denials would push revenue further out and raise the cost of capital for the whole sector.
Energy Transfer's next earnings report, due November 3, 2026, will be the first opportunity for management to update investors on the South Mississippi commitment and the status of its broader data-center backlog. The company's guidance on 2026 growth capital expenditure — currently anchored by the Hugh Brinson and Desert Southwest projects — will show whether the pipeline is moving from announcement to dirt-moving.
The Bottom Line
Energy Transfer's $1.6 billion commitment is more than a pipeline approval; it is a vote on the shape of the next decade of American energy demand. The structural case — that AI data centers and LNG exports have created a new, durable load class that only interstate pipelines can serve — is strong, and the company has insulated itself by letting customers fund nearly half the project. But the trade is not one-sided: the demand forecasts underpinning it carry wide error bars, the stock has already rallied on the narrative, and a slowdown in data-center final investment decisions would leave expensive steel waiting for volumes that arrive later than expected.
The takeaway: this is the market pricing the deficit, not a cyclical dip — the bottleneck has moved from chips to electrons, and now to the pipes that carry the gas to make them. Investors should underwrite the contracted cash flows and treat the expansion optionality as exactly that: an option, not a guarantee.
Explore more exclusive insights at nextfin.ai.
