NextFin News - The AI buildout is no longer just a chip and server story. It is becoming a pipeline story, because the fastest way to feed new data-center load is increasingly to move more gas to places where firms can turn that fuel into firm power. Energy Transfer said its Transwestern Pipeline Desert Southwest expansion will add natural-gas capacity for Arizona and New Mexico, a project it described as driven by population growth, high-tech industry demand and data-center expansion. The timing is telling: the company expects the project to be in service in the fourth quarter of 2029, while the near-term power-demand math keeps getting bigger.
EIA now expects U.S. electricity use to grow by 1% in 2026 and 3% in 2027, and it says the increase is being driven by large computing centers. EIA also launched pilot surveys this year to measure data-center energy use, a sign that the government is still trying to map a market that is changing faster than its historical datasets were built to track. Goldman Sachs Research says U.S. data-center power demand could rise from 31 gigawatts in 2025 to 41 gigawatts in 2026 and 66 gigawatts in 2027, with data centers’ share of peak summer demand rising to 8.5% by 2027. Morgan Stanley has separately estimated a U.S. power shortfall for data centers of 44 gigawatts through 2028 before time-to-power workarounds.
The result is a new ordering of the energy system. Gas pipelines are not being asked to replace the grid, but to narrow the gap between the speed of digital demand and the slower pace of transmission, interconnection and large-scale generation. That gap matters because it changes who captures the economics. If the binding constraint is fuel delivery into fast-growing load pockets, then the value migrates toward operators that can place molecules closest to new generation and data-center corridors. If the constraint is solved instead by behind-the-meter generation, batteries or other time-to-power approaches, the midstream benefit narrows.
That tension is why the current surge in pipeline announcements looks less like a cyclical burst and more like a structural revaluation of infrastructure. Cyclical demand surges normally fade once inventories, financing or commodity prices normalize. Here the driver is a multi-year technology deployment that changes the level of electricity demand, not just its timing. EIA’s forecast of four straight years of electricity-demand growth and Goldman Sachs’ two-year jump from 31 gigawatts to 66 gigawatts point in the same direction: AI demand is not a weather event for the grid. It is a new base load.
Why The Pipeline Link Matters More Than The Headline Number
The key mechanism is not simply that data centers need more electricity. It is that they need power on a timetable that traditional infrastructure cannot always meet. A hyperscale campus can move from site selection to financing to build in a matter of months, but new transmission, interconnection and large utility-scale generation can take years. In that mismatch, natural gas has an advantage because it can be dispatched reliably and paired with assets that already understand how to move and contract around firm load.
Energy Transfer is leaning directly into that mechanism. In its June 2026 investor presentation, the company said it had more than 6 billion cubic feet per day of contracted pipeline capacity and highlighted agreements tied to Oracle data centers and AI campus projects. It also said its Hugh Brinson Pipeline was in commercial service and that it was working toward placing the full phase I capacity of 1.5 billion cubic feet per day in service. Those numbers matter because they show the company is not just talking about demand; it is trying to lock in fee-based volume before the load fully arrives.
The Transwestern Desert Southwest project follows the same logic, but in a different geography. The company said the expansion will serve utilities and energy providers in Arizona and New Mexico, and it expects the project to be fully subscribed after the open season. The expected cost of about $5.3 billion, including $0.6 billion of allowance for funds used during construction, shows that this is not a marginal add-on. It is a capital decision large enough to require confidence that the load curve will outlast a single market cycle.
That is where the structural thesis becomes stronger than the cyclical one. A cyclical story tends to come with a visible inventory overhang or a financing window that closes. This buildout is driven by an underlying reconfiguration of computing architecture. AI training and inference workloads require dense, reliable power, and the more those workloads move from pilots into production, the more the system needs firm electricity on tap. The demand is therefore cumulative. A new data hall does not erase the previous one; it adds to the base.
Energy Transfer management framed the company’s own outlook in similarly durable terms. On its second-quarter earnings call, it said adjusted EBITDA would be between $18.8 billion and $19.1 billion for the full year, and it said the Hugh Brinson project was already in commercial service. The point is not the earnings line itself. It is that pipeline owners now see AI-linked load as contractible enough to justify long-dated asset buildouts.
“EIA expects U.S. electricity use to grow by 1% this year and 3% in 2027,” the agency said on Jan. 13, 2026, adding that the increase would mark the first time since 2007 that power demand has risen for four years in a row and the strongest four-year growth period since 2000.
That sentence is the cleanest summary of the regime shift. A grid that is growing for four straight years because of computing demand is not dealing with a transient weather-driven spike or a one-off industrial rebound. It is dealing with a new source of load that is more persistent than previous demand shocks.
What The Market Is Already Pricing - And What It Is Not
The market is clearly pricing the idea that AI will require more power. That part is already conventional wisdom. The more interesting question is how far the economics travel down the chain. The first-order winners are easy to name: gas pipeline operators, gas producers and equipment makers that help connect fuel to load. But the second-order effects can be more selective. If the best solution for a region is a mix of gas-fired generation and new pipeline capacity, then the pipeline owner with the right geography and contracts can enjoy recurring fee-based revenue. If the better solution is on-site power, batteries, fuel cells or non-gas alternatives, the pipeline’s role shrinks even if the data-center buildout keeps accelerating.
This is why the story should not be read as a blanket endorsement of midstream assets. It is a test of which part of the power stack can solve time-to-power fastest. The more local and immediate the power requirement, the more valuable the pipeline becomes. The more the market can satisfy load through distributed generation or non-pipeline solutions, the less of that demand spills into long-haul transport.
That second-order view also explains why the obvious bullish conclusion can be too simple. A lot of the current enthusiasm assumes that every extra megawatt for AI eventually turns into a higher transport fee somewhere in the chain. That may be true in some corridors, but not everywhere. The real benefit sits with the assets that control the bottleneck, not with the entire sector by default.
The strongest counter-thesis is that this is still a cyclical trade hiding inside a structural narrative. On that reading, today’s pipeline projects are being announced in front of demand that is already obvious, and the market may be overpaying for the timing advantage. Project delays, rising construction costs, or a faster-than-expected shift toward alternative power solutions could leave some of the economics stranded. The counter-argument is serious because energy infrastructure has a long history of overbuilding when a new theme attracts capital.
But the falsifying signal is also clear: if EIA, Goldman Sachs Research and other load forecasters begin cutting their 2027-2028 demand assumptions, or if major data-center commitments fail to convert into physical buildouts, then the structural case weakens. Short of that, the burden of proof sits with the skeptics, because the demand curve itself is still climbing.
What To Watch From Here
In the short term, watch whether pipeline operators can move from announcement to open season, financing and in-service without a meaningful delay. The current theme only matters if the assets reach completion before the load catches up. That makes permitting, construction milestones and contract signings more important than the headline size of any one project.
Over the medium term, the key question is whether gas remains the most efficient bridge between AI load growth and reliable power. If it does, operators with existing corridor access and firm contracts should keep pulling in the economics. If the industry shifts toward behind-the-meter generation, storage or other time-to-power solutions, the pipeline trade becomes narrower and more geographically selective.
Over the long term, the broader implication is that electricity infrastructure is being rebuilt around a technology demand shock rather than a traditional industrial cycle. That favors assets with contractual visibility and physical leverage over the load pocket. It also means that companies controlling fuel delivery, generation or interconnection can become more valuable simply because the old grid is too slow for the new economy.
The base case is continued buildout of gas-linked capacity as AI demand keeps outpacing the power system’s ability to respond. The upside case is that operators secure more long-dated contracts and place more capacity into service faster than expected. The downside case is that project delays and alternative power solutions absorb much of the incremental demand before it reaches the pipeline network.
The clearest takeaway is that AI is no longer just raising electricity demand. It is changing which infrastructure can still be built fast enough to matter.
This is not a temporary spike in power demand. It is the market learning that megawatts, not model size, are becoming the real constraint.
