The Water Problem: The Midstream Nobody Renders
Written by Bruno P. Allaire
Every AI campus announcement renders the turbines and the buildings. Nobody renders the pipelines, the laterals, or the water handling. That unglamorous infrastructure is where projects actually live or die, and where the operators compound quietly.
Share this perspective
What the Rendering Leaves Out
Open almost any announcement for an AI campus and you will see the same picture. A long low building, a field of cooling units, a substation, and in the background, some suggestion of generation. What is never in the picture is the part I spent much of my career building: the pipe in the ground, the water handling, the disposal capacity, the laterals that connect a site to a system that was planned decades before anyone thought about inference workloads.
That omission is not cosmetic. It is where the schedule actually breaks. A campus needs gas delivered at a pressure and a volume, and it needs water for cooling and for the people who work there, and it needs somewhere for the water it produces or rejects to go. Each of those is a permit, a contract, and a physical asset, and each of them was built for a different purpose by people who never expected this demand. The unglamorous midstream is the difference between a project that energizes and a project that is a rendering with a press release attached.
I have spent enough years on the midstream side to have a bias, so let me name it. The exciting part of this buildout is the compute. The durable part is the infrastructure that makes the compute possible, and infrastructure is won by whoever controls the boring assets before the demand arrives. That is not a new lesson. It is the lesson of every energy cycle I have worked through, and it is being replayed at a scale that has caught the industry flat-footed.
Water Is a Permit Problem First
In oil and gas, water is a byproduct you have to manage, and the management is a permitting exercise before it is an engineering one. Produced water comes up with the hydrocarbons, and in many fields the volumes exceed the oil by multiples. It gets separated, treated, and either recycled into operations or disposed of in permitted disposal wells. Every one of those steps is regulated, and the disposal capacity is a finite, permitted quantity attached to specific wells with specific UIC permits.
That matters for the AI buildout because the same regulatory architecture now sits in front of a data center's water needs. A campus needs an industrial water supply, which means a withdrawal permit or a municipal contract, and it needs a discharge or reuse plan for cooling water, which means a different permit with a different agency on a different timeline. In water-stressed jurisdictions, and the most attractive siting jurisdictions for a variety of reasons tend to be exactly those, the water permit can be the longest lead item on the entire schedule.
The practical consequence is that water is a siting constraint that presents as an engineering detail. A sponsor who budgets for cooling towers and assumes the water will be there is running the same diligence gap as the sponsor who assumes the gas will be there. Both are assuming that a resource with a permitting queue and an incumbent user base is somehow available on demand, because the model needed it to be.
The Midstream Gap
Here is the structural feature of this market that I think is least understood outside the industry. The midstream network in the United States was not built to serve data centers. It was built to serve production, gathering gas from wells and moving it to processing and interstate takeaway, or to move water from separators to disposal. Those networks are dense in the producing basins and thin everywhere else, and a data center load is a point demand that requires a lateral, a metering station, and often compression, none of which exist at the site until someone builds them.
That creates an unusual opportunity for the operators who already own the pipe. A midstream company with an existing right of way, an existing permit posture, and an existing relationship with the pipeline downstream can build a lateral in a fraction of the time a greenfield developer can, and it can do so while capturing the transportation fee rather than watching someone else capture it. The value of an existing right of way in a constrained county is not a figure that appears in a rendering. It is the asset.
The water side mirrors it. An operator with permitted disposal capacity, treatment equipment, and the operational history to run both is sitting on something a campus developer needs and cannot quickly replicate. That is not a licensing business in the sense of a technology license. It is a physical, permitted, geographically bounded business with real barriers, and barriers are what compound.
Why the Unglamorous Compounds
The operators who compound in this cycle are not the ones with the best story. They are the ones who control the three or four physical inputs every project needs and nobody wants to build: fuel delivery, water supply, water disposal, and the right of way that connects the site to the system. Each of those is small relative to the campus. Each of them is decisive for the campus, and each of them is worth more to the second campus than the first, because the second one can be served off the same infrastructure.
That is the compounding mechanism, and it is the same one I have watched work in every energy cycle. You build a lateral for one customer and you size it for three. You permit a disposal well for your own production and discover it is the only permitted capacity within fifty miles of a new load. You buy an existing midstream asset at a cyclical low because the cash flow is unremarkable, and then the demand for the thing it moves changes the value of the asset permanently.
The reason this stays quiet is that none of it photographs. There is no ribbon cutting for a water handling facility, and the operational excellence that makes one work is invisible even to the people who benefit from it. But the returns are real, and they are unusually durable, because the barrier is not capital. It is time, permits, and relationships in a specific county where nobody else has them.
What to Diligence Before You Site
For a sponsor or an investor evaluating a project, the water and midstream work should be part of the siting decision, not part of the pre-construction scramble. The sequence is short. Which permitted disposal capacity exists within economic hauling distance, and who owns it? Which treatment path is available, at what cost, and does the treated water have a reuse market or is it a disposal obligation? Where does the industrial water supply come from, is it a withdrawal permit or a contract, and what is the renewal risk?
Then the pipeline questions from the other direction. Is there an existing right of way that reaches the site, and who holds it? What is the pressure and volume available at the nearest delivery point, and what would it cost to extend? Does the local system have the compression and treatment to accept the quality of gas you intend to deliver? Each answer changes the project's capital cost, and the sum of them frequently changes whether the project should be built at all.
I would add one more item that is easy to miss. Ask who in the county has done this before. The permitting agency, the pipeline operator, the disposal well owner, the county engineer: a project in a jurisdiction with a recent history of similar approvals is a different project from one that will be the first of its kind. That institutional memory is not on any map, and it is worth more than a favorable power price.
The Quiet Business
The AI buildout is going to be financed, built, and eventually corrected, and when the correction comes it will separate the projects that owned their infrastructure from the ones that leased a promise. The campuses with contracted fuel, permitted water, and a midstream connection they control will still be operating. The ones that were assembled around a rendering will be repriced, and the infrastructure underneath them will move to whoever can actually run it.
I think that is why the water question is worth writing about now, while the demand is still arriving and the mistakes are still cheap to avoid. The demand is not in doubt: forecasters expect data center electricity consumption to grow by multiples of its current share by the end of the decade1. What is in doubt is whether the projects being announced have done the work on the parts that never make the slide. The midstream is where that work happens, and it is where the operators who have been doing it for twenty years have a structural advantage that will not be competed away quickly.
Compute scales. Power does not, and neither does pipe. The people who understand that are not usually the people writing the announcements, which is precisely why the opportunity exists.
Notes & Sources
- 1.EPRI projects that data centers could consume between 4.6% and 9.1% of all US electricity by 2030, up from roughly 4% today. Electric Power Research Institute, "Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption," 2024