AI Has a Pipeline Problem: Part 1
Richard Kinder, data centers, and the physical constraint behind the intelligence boom
By Bryan J. Kaus
“We shape our buildings; thereafter they shape us.” — Winston Churchill
AI feels weightless from the user interface.
A prompt goes in. An answer comes out. A model writes, codes, summarizes, searches, analyzes, drafts, edits, translates, and reasons in seconds. The experience is digital, frictionless, and almost abstract.
The infrastructure behind it is anything but.
Behind the intelligence is land. Power. Gas. Wires. Water. Steel. Cooling. Transformers. Substations. Turbines. Pipelines. Permits. Construction crews. Financing markets. Interconnection queues. Skilled labor. Regulatory approval. Community acceptance. And time.
That is the part of the AI story the market is still working through (and expectations will temper to reality).
The last piece I wrote on AI focused on capital governance: the risk that when everything feels possible, companies begin treating possibility as priority. AI has become large enough that it is no longer just a technology story. It is a capital-intensity cycle. A balance-sheet story. A credit story. A management discipline story.
This is the next layer.
AI may be built in code, but it will scale through infrastructure.
And infrastructure has physics.
The Issue
Richard Kinder has been one of the clearer voices on this point.
That matters because Kinder is not selling an AI model. He is not underwriting a software multiple. Kinder Morgan is a physical infrastructure company. It moves energy. It sees demand through contracts, utilization, bottlenecks, expansion projects, creditworthy customers, and whether someone is willing to commit capital for deliverable capacity.
That vantage point is useful.
On Kinder Morgan’s first-quarter 2026 earnings call, Kinder said he had looked back at his own prior comments about natural gas demand and found that, “in almost every case,” the projections he made turned out to be understated. His explanation was straightforward: demand for natural gas, driven primarily by LNG feedgas and increased utilization of natural gas for electric generation, has grown faster than expected. He then added that recent events had made the outlook even more positive.
The data point that jumps out is electric generation.
Kinder cited S&P Global Market Intelligence reporting that utilities plan to add 153 gigawatts of gas-fired generation capacity over the next several years, primarily to serve data centers, with the bulk of that coming online by 2030. He noted this was twice the estimate from the same group a year earlier and reflected plans to build about 210 additional natural-gas-fired facilities. Kinder Morgan’s own forecast now sees total U.S. gas demand reaching approximately 150 Bcf per day in 2031, growth of about 27% from this year. His summary was simple: “the natural gas story has legs.”
What sits behind that headline is more interesting.
On the same call, Kinder Morgan’s president Dax Sanders told analysts that the company is in various stages of development on projects serving more than 10 Bcf per day of natural gas demand in the power generation sector and over 3 Bcf per day in the LNG sector. These are not booked projects in the $10.1 billion backlog. They are what is moving through the operator’s commercial pipeline behind the headline number.
That is the deeper read.
Not a slogan.
Not a pitch deck.
A demand forecast tied to generation, data centers, LNG, and pipeline capacity, with another layer of identified opportunity sitting beyond what has already been formally committed.
The important nuance is that this is not a claim that every announced data center will be built. That would be the wrong conclusion. Some projects will not get built. Some will be delayed. Some will be repriced. Some will fail on interconnection. Some will fail on permitting. Some will fail because financing conditions change. Some will fail because the customer economics of AI take longer to prove out than the capital spending assumed.
But not every project has to be built for the infrastructure requirement to be enormous.
That is the point.
A discounted version of the AI and data center buildout can still create a massive need for power generation, gas deliverability, pipeline capacity, transmission, substations, cooling systems, construction labor, and capital.
The opportunity is real.
So is the constraint.
The Physical Chain
The AI economy is creating a surge in expected power demand. That part is no longer theoretical.
American Electric Power recently raised its five-year capital investment plan to $78 billion, up from $72 billion, citing demand growth in key states including Indiana, Ohio, Oklahoma, and Texas. AEP signed 7 gigawatts of new large-load agreements during the first quarter, primarily in Ohio and Texas, bringing total expected incremental load to 63 gigawatts by 2030, with AEP Texas accounting for 41 gigawatts of those commitments.
That distinction matters.
Demand is one thing.
Deliverable power is another.
A data center does not run on market narrative. It runs on electrons. Those electrons have to be generated, transmitted, delivered, stabilized, cooled, backed up, and paid for.
That is where the infrastructure chain matters.
Data centers require power.
Power generation often requires gas.
Gas requires pipelines.
Pipelines require permitting, rights-of-way, compression, steel, construction, regulatory approval, and customers willing to sign contracts.
Transmission requires lines, towers, substations, transformers, land, interconnection, permitting, and cost recovery.
Cooling requires water, equipment, planning, and local acceptance.
Financing requires confidence that the demand, counterparty, contract structure, and return profile are real.
Every link matters.
Kinder’s Warning, and Kinder’s Opportunity
Richard Kinder’s argument is grounded but generally bullish.
That is what makes it useful.
He is not telling the market to chase every AI-adjacent story. He is saying the underlying demand has grown faster than expected, and the midstream sector, underpinned by long-term throughput agreements with investment-grade customers, offers a comparatively low-risk way to participate in that growth.
Kinder’s framing on the call was direct on positioning: he described Kinder Morgan’s pipeline network as “enormously advantaged by the sheer size and location” of its assets, located in the areas where gas demand is growing dramatically. The strategy he described is to expand and extend existing assets in an aggressive but disciplined manner, complete projects on time and on budget, and finance growth primarily with internally generated cash flow while maintaining a strong balance sheet.
That is an operator’s answer.
It is also an investor’s answer.
Own the scarce asset.
Contract with creditworthy customers.
Avoid speculative overreach.
Fund growth without breaking the balance sheet.
Execute.
That is very different from chasing the broad label of “AI infrastructure.” And there are so many out there simply chasing…
Kinder Morgan’s hard numbers support the case. In Q1 2026, the company reported utilization on its five largest gas pipelines exceeded 90%, up from 74% in 2016. Its project backlog reached $10.1 billion, with approximately 92% tied to natural gas projects and nearly 60% supporting power generation and local distribution company demand. New projects added in the quarter alone totaled $375 million, including three deals tied directly to data center load. The footprint behind those numbers is 78,000 miles of pipeline and 136 terminals, much of it sitting in the corridors where demand is moving fastest.
There is also a less-discussed competitive layer.
Natural gas storage as a structural differentiator. The company holds more than 700 Bcf of storage capacity, with additional expansions under evaluation. As data center load, LNG cycling, and weather-driven demand grow, the ability to inject and withdraw gas quickly stops being just inventory and starts being part of the system’s reliability infrastructure. Storage becomes a scarce resource, not a commodity service. That is the kind of asset detail most market commentary misses.
Again, that is not a promotional AI number.
It is an infrastructure signal.
A utilization problem.
A backlog.
A development pipeline beyond the backlog.
A storage moat.
A tightening system.
A customer base asking for deliverable energy.
Kinder Morgan also highlighted specific projects tied to this demand. Its Creekside Lateral project is designed to serve growing power generation, industrial, and data center demand in Central Texas and is supported by binding long-term contracts. The company is also advancing larger projects such as the Trident Intrastate Pipeline, intended to move significant volumes from Katy, Texas toward the industrial corridor near Port Arthur.
That is the real-world version of the AI story.
Pipe in the ground.
Contracts.
Customers.
Storage.
Steel.
Time.
The Buildout Is Real
The broader market evidence points in the same direction.
S&P Global’s Regulatory Research Associates now forecasts approximately $1.3 trillion of aggregate capital expenditures for U.S. energy utilities between 2026 and 2030, as companies seek to modernize infrastructure and add generation capacity in response to heightened energy demand from large-load customers, particularly data centers that support AI, digital services, and cloud infrastructure. Data centers and other large industrial loads are expected to add 374 TWh of energy demand and over 45 GW of peak load through 2035, according to analysts at S&P Global Energy CERA.
That turns AI into a utility-capex story.
It is not simply about model performance.
It is about financing the physical layer of the digital economy.
Morgan Stanley has estimated that global data center capacity could increase by a factor of six over the next five years, with spending on data centers and hardware alone reaching roughly $3 trillion by the end of 2028. Roughly half of that funding is expected to come from hyperscaler cash flows, with the rest financed through outside channels of the credit markets.
Meta’s planned El Paso data center illustrates the scale. Meta is working with Morgan Stanley and JPMorgan Chase on a roughly $13 billion financing package for the project. Meta had previously increased its investment in the planned El Paso AI data center by more than sixfold to $10 billion, aiming to reach 1 gigawatt of capacity ahead of the facility’s projected 2028 opening.
One gigawatt is not a rounding error.
It is a power-plant-scale requirement attached to a digital business model.
That is the shift.
AI is pulling the physical economy forward: utilities, pipelines, gas generation, substations, transmission lines, construction, turbines, engineering firms, project finance, land development, equipment suppliers, cooling systems, water infrastructure, and skilled trades.
Williams is seeing the same kind of demand from the pipeline side. In Q1 2026, Williams upsized its Transco Power Express project, increasing capacity to 750 million cubic feet per day, aimed at serving rising energy demand from AI-driven data centers in Virginia. The company is on track to deliver 2026 adjusted core earnings at the higher end of its forecast range, supported by data center and LNG export demand.
Exelon is another current example. The company raised its four-year capital expenditure plan to $41.7 billion, projecting 7.9% rate base growth. Executives said the data center pipeline is increasingly backed by FERC-approved Transmission Security Agreements, which have now secured approximately $1 billion of collateral. Transmission rate base is now projected to grow at 16% through 2029.
The pattern is becoming visible.
Utilities are increasing capital plans.
Midstream companies are adding pipeline capacity.
Hyperscalers are arranging large project financings.
Developers are signing long-term leases.
Construction and equipment ecosystems are being pulled into the cycle.
This is not merely a technology trade.
It is an industrial buildout.
The Point Taken
AI may feel weightless at the user interface. The system that powers it does not.
Utility capital plans are rising. Midstream backlogs are expanding, and operator commercial pipelines sit deeper still. Financing packages are being assembled at scales the industry has not seen in a generation. Construction labor, equipment, transmission, gas deliverability, and storage are all being pulled into the same cycle.
The demand is real.
The physical chain is real.
The system is being asked to deliver more than it has in decades.
That is the part of the story the market is still working through.
The harder questions come next. How the buildout will actually be funded. Where the local constraints will bind. Which operators capture durable value, and which projects end up as stranded monuments to a cycle that moved faster on paper than in the field.
Stay tuned for part two next week…