Closed loop? It's not that simple.

I'm an advocate for responsible deployment of AI and compute. I build software for a living and I run inference hardware in my own house. So take this from someone on the industry's side of the argument: the water concern here is real, and it is not going to solve itself. The reassurance you'll hear is that modern chips don't need evaporative cooling. True - and irrelevant, because nobody is obligated to put modern chips in the building. Depreciated hardware is the cheapest compute there is, and it comes with the cooling it was designed around.

Here's the problem. "Closed loop" describes the loop. It does not describe the building.

A modern rack runs liquid straight to the chip. That coolant circulates in a sealed loop, picks up heat, and carries it out to the edge of the facility. Nothing evaporates. Nothing is consumed. Every word of the marketing is true, right up to the point where that heat has to actually leave.

Because it does have to leave. The loop can be perfectly sealed and the heat still has to go somewhere, and the equipment at that boundary is where the water question actually lives. Put a dry cooler there - essentially a very large radiator - and the heat goes into the air. No water consumed. That's the real thing.

Or install a dry cooler with adiabatic assist. It looks nearly identical. It sits in the same place in the same diagram. But when ambient temperature climbs, it starts spraying water across the coils or wetting media in front of them, and that water evaporates into the airstream to boost capacity. That is evaporative cooling. It is consumptive. And the loop is still closed, so every claim made in the public meeting stays technically true.

Notice the timing. These hybrid systems run dry most of the year and switch to wet mode on the hottest days - which in Yellowstone County are the same days the river is lowest and irrigators are already fighting over the last of it. The system is designed to consume water at precisely the moment the community can least afford it. Annual averages will look wonderful in the sustainability report.

So the standard cannot be "closed loop." It has to be written as an outcome, and it has to name the hardware it excludes:

Zero consumptive water use for heat rejection. Dry cooling only. No evaporative towers. No adiabatic pre-cooling, spray, or wetted media. No hybrid units capable of switching to a water-consuming mode at any temperature threshold. Independently verified and reported annually. Any change requires reopening the agreement.

That last clause matters more than it looks. Without it, a facility complies on day one and quietly installs assist modules in year four when a hot summer squeezes their capacity.

And I want this at the state level, not left to individual towns. If Montana handles this parcel by parcel, every community negotiates alone against a developer who has run this play in a dozen other states and knows exactly which town has a part-time planner and no engineering staff. The towns with the least capacity to push back will end up with the wet cooling. That is not a hypothetical - it's the entire business model of siting.

The counterargument will be cost. Dry cooling uses more electricity, and adiabatic assist saves them money. That is true, and it is not our problem. The extra power is a line on their operating statement. The water is the only one we have. A company asking to spend our river to lower their utility bill should be told no, and should be told it in statute rather than in forty separate council chambers.

Direct-to-chip liquid runs warm - the supply water doesn't need to be chilled the way old air-cooled halls did which means dry cooling works fine for a modern facility, especially in a Montana climate. The technology to do this right exists and is shipping today. Anyone insisting they need the wet option is telling us something about their intentions, and we should listen.

Build here. Pay taxes here. Hire here. Just don't drink our rivers, pollute our aquifers, and lie to our citizens.