The Data Center Crisis: Why the Future of AI is Moving to 8,000 Feet
The Data Center Crisis: Why the Future of AI is Moving to 8,000 Feet
As the power grid hits a breaking point in Northern Virginia, the answer to AI's energy crisis is hiding in plain sight at 8,000 feet
#AIDataCenters #Power #Cooling #Latency #MountainWest
First, What's the Problem?
AI runs on computers. A lot of computers. Massive buildings full of them, running 24 hours a day, seven days a week, generating enormous amounts of heat.
That heat has to go somewhere.
Getting rid of that heat — cooling — costs almost as much energy as the computing itself. For every dollar of electricity a data center spends doing useful AI work, it spends roughly another 55 cents just moving heat around. That's not a rounding error. At the scale of modern AI infrastructure, that's billions of dollars a year being spent on the equivalent of a really, really big air conditioner.
So the data center industry has three giant problems it needs to solve at the same time:
Where do we get enough power? AI data centers are power-hungry on a scale that's straining electrical grids across the country.
How do we keep everything cool? You can't just pile GPUs into a building and walk away. They'll melt. The cooling bill alone can define whether a facility is profitable.
Where do we even build? The obvious places — Northern Virginia, Phoenix, the Bay Area — are running out of land, grid capacity, and patience from local governments.
There are some wild ideas floating around for how to solve these problems. Orbital data centers in space. Underwater data centers on the seafloor. We'll talk about both. But the most elegant answer doesn't require a rocket or a submarine.
It requires a map and a willingness to look west.
Here's the Idea: Go High, Go West
The western slope of the Rocky Mountains — think western Montana, Idaho, Wyoming, western Colorado — solves all three problems at once. Not through exotic engineering. Through geography.
Here's how.
Problem 1: Cooling. Nature Already Solved This One.
Air conditioning is expensive. Free cold air is free.
At elevation in the Mountain West, the air is cold for most of the year. Not "pleasant autumn breeze" cold. Legitimately cold. A north-facing slope at 8,000 feet in Montana can be cold enough in July that you need to heat a building — not cool it. There's a person in Montana who built a home in a canyon on the north side of the mountains. He has no choice but to heat his house in summer. The mountain blocks the sun, cold air rolls down from the snowfields above, and the building loses heat instead of gaining it.
That's exactly what you want for a data center.
When outdoor air is cold enough, you can just pull it through the building to cool the servers. No compressors, no refrigerant, no giant chillers humming around the clock. Engineers call this "free cooling" or "air-side economization." The name is accurate — it's genuinely close to free.
Most data center locations can only use free cooling for a few weeks out of the year. The Mountain West at elevation can use it for most of the year. That changes the economics dramatically.
And here's a bonus most people don't think about: the air is dry. Not desert dry — not Death Valley — but consistently low humidity. That matters for hardware. Moisture corrodes metal, causes condensation, and shortens the life of expensive equipment. Microsoft ran an experiment called Project Natick where they sealed servers in a nitrogen-filled container and dropped it on the seafloor off Scotland for two years. The failure rate was one-eighth of comparable land-based servers. Why? Stable atmosphere, low corrosion, nobody bumping into things. The Mountain West doesn't give you nitrogen atmosphere, but it does give you dry, clean air — and that extends hardware life for similar reasons.
Problem 2: Power. You Want to Be Close to Where It's Made.
Electricity doesn't teleport. It travels through transmission lines, and the farther it travels, the more you lose along the way — typically 5 to 8 percent of the electricity generated simply disappears as heat in the wires before it reaches the plug. More importantly, the most popular data center markets — Northern Virginia especially — are hitting genuine gridlock. The grid operator there, called PJM, is telling new customers to expect waits of five to seven years just to get a power connection approved. Not five to seven months. Five to seven years. Companies are reserving spots in that queue right now for data centers they haven't started designing yet.
The western United States runs on a separate electrical system from the east called the Western Interconnection. It stretches from the Rockies to the Pacific, from Canada to Mexico. It carries about 250 gigawatts of generating capacity, and it's growing fast — almost 90 percent of all new power capacity added to the Western grid in 2023 came from solar, wind, and battery storage.
The Mountain West sits right next to the generation sources that feed this grid. Wyoming wind. Montana hydro. Utah and Nevada solar. You're not at the end of a long transmission line from where the power is made — you're at the beginning of it. That means cheaper power, more reliable supply, and a shorter queue to get connected.
There's also a sustainability angle that matters to the companies building these facilities. AI firms are under real pressure to hit net-zero goals. When a data center in Wyoming buys power from a Wyoming wind farm on the same grid, those electrons are physically the ones spinning the servers. That's a much cleaner story than buying carbon offsets for a facility in a coal-heavy grid region and calling it green.
The Eastern grid, where most of today's data centers are being built, is congested and getting worse. The Western grid has room. That gap is a real opportunity for the next wave of infrastructure investment.
Problem 3: Will It Be Slow? No. Here's Why.
A reasonable question: if the data center is in rural Montana, won't responses to users in New York or Los Angeles be slow?
Not really. Here's the thing — fiber optic cable runs everywhere, including straight through the Rockies. The I-80 and I-90 highway corridors are major transcontinental fiber routes. Cheyenne, Wyoming is already a real crossroads for fiber traffic crossing the country — it's not a tech hub in the popular imagination, but the cables don't care about the imagination. Denver is a legitimate internet exchange point. Salt Lake City has solid connectivity. A data center in western Montana or Wyoming can reach Seattle, San Francisco, or Los Angeles in single-digit milliseconds over fiber. That's the same ballpark as a data center in suburban Virginia serving someone in Washington D.C.
Now compare that to the exotic alternatives. A data center in orbit — which is an actual pitch being made in the industry right now — has to deal with the basic physics of distance. Even at low Earth orbit, the round-trip signal time is at minimum 20 to 40 milliseconds, before you add satellite handoffs and ground station routing. For AI applications that involve thousands of back-and-forth calls — like an AI coding assistant — that latency adds up fast and noticeably degrades the experience.
Mountain West on the ground, on fiber: no latency problem. Mountain West in orbit: real latency problem. The ground wins.
What About Those Wild Alternatives?
Let's be fair to the other ideas, because they're real engineering programs with real smart people working on them.
Orbital data centers — the idea of putting compute in space — actually solves power and cooling in an elegant way. You get near-unlimited solar power and vacuum cooling that requires no refrigerant whatsoever. The catch: you still have to get hardware up there (expensive), you can't fix it when it breaks (really expensive), and the latency makes it unsuitable for serving real-time AI responses to users. It makes sense as a place to run long AI training jobs where latency doesn't matter and raw compute hours do. It does not make sense as the server that answers your question.
Underwater data centers — Microsoft actually tried this. Project Natick was a real experiment, not a PowerPoint. They submerged 855 servers in a sealed container off the coast of Scotland and left them for two years. The hardware lasted remarkably well — eight times better than comparable land-based servers. Seawater is an excellent, free heat sink. The problem is power. Getting electricity to the seafloor requires expensive submarine cables, creates single points of failure, and doesn't get any cheaper at scale. Cooling: solved. Power: not solved. Microsoft ultimately shut the project down.
The Mountain West addresses cooling and power and latency without requiring a launch vehicle or a submarine. That's the comparison that matters.
The Land Situation: More Than You'd Think
Here's something that doesn't get talked about enough: the popular data center markets are expensive and complicated in ways that go beyond just power and cooling.
Land in Loudoun County, Virginia — the data center capital of the world — has gotten genuinely crazy. Competition for grid-connected parcels has driven prices to levels that were unimaginable ten years ago. Phoenix has absorbed enormous investment but is now bumping into both land scarcity and water supply concerns. (Evaporative cooling towers — the most common large-scale cooling approach in hot climates — use millions of gallons of water a year. In the desert Southwest, that's a real problem. Mountain West air cooling uses essentially no water.)
Rural Montana, Idaho, Wyoming, and Colorado have abundant land, motivated local governments looking for tax base and employment, and a permitting environment that is simpler than California or the Mid-Atlantic by a wide margin.
One more thing: earthquake risk. The Mountain West on the eastern side of the Rockies sits on stable geology. The fault line risk that makes the Pacific Coast and parts of the Southeast genuinely concerning for long-lived infrastructure doesn't apply the same way to a well-chosen site in central Idaho or western Montana.
"But Who's Going to Work There?"
This is the objection that stopped earlier versions of this conversation, and it's fair. Data centers need people — technicians, engineers, operators. Those people historically wanted to live in cities. Rural Montana is not a city.
Two things are changing that calculation.
First, the staffing model for modern data centers is already lean and getting leaner. Automation, remote monitoring, and predictive maintenance have been reducing the ratio of human operators to computing capacity for years. The next wave of AI-driven operations tools — systems that diagnose problems, route maintenance tasks, and handle routine operations without human involvement — is pushing that ratio even further. A modern high-efficiency facility doesn't need hundreds of people on site. It needs a small, skilled team for hands-on work, backed by remote management for everything else.
Second, the Mountain West has university towns that are quietly becoming magnets for people who want the knowledge-economy lifestyle without the cost and congestion of Seattle or San Francisco. Bozeman, Montana. Moscow, Idaho. Laramie, Wyoming. Grand Junction, Colorado. These places have engineering schools, lower cost of living, and access to outdoor recreation that increasingly attracts exactly the kind of people who work in technical operations. Montana State University in Bozeman has a growing computer science and engineering program. The workforce pipeline is real, even if it doesn't look like Silicon Valley.
The workforce objection was valid for a while. It's dissolving.
Okay, But Here Are the Real Objections
Every good thesis has weak spots. This one has two worth being honest about.
The air is thinner up there. At 8,000 feet, the atmosphere is less dense than at sea level. Thinner air carries less heat, which means fans have to spin a little faster to move the same amount of cooling through a server rack. This is a real effect. It slightly offsets the free-cooling gains. Data center engineers know about it and design for it — larger heat sinks, adjusted fan curves, thermal modeling that accounts for altitude. It doesn't erase the cooling advantage, but it does mean "free cooling at elevation" isn't quite as simple as "cold air goes in, hot air goes out." The net benefit still strongly favors the Mountain West. But anyone who says altitude has zero cost isn't being fully straight with you.
Getting equipment there is harder. Building a data center in rural Montana means trucking thousands of heavy GPU racks, specialized transformers, and high-voltage electrical gear over mountain passes to locations that are not currently on any established data center logistics route. The contractors who wire hyperscale facilities are concentrated in Northern Virginia, Phoenix, and the Pacific Northwest. High-voltage electricians with hyperscale experience are not waiting in Bozeman for the phone to ring.
This is the most legitimate friction point in the entire thesis. It's not a dealbreaker — remote industrial construction happens all the time, and the Mountain West has a long history of building large energy infrastructure in challenging terrain. But it adds cost and timeline to the build phase that a greenfield developer needs to plan for honestly. The operational advantages compound over decades. The logistics challenge is front-loaded. That's a tradeoff worth naming clearly.
So Why Isn't Everyone Already Doing This?
Honestly? Because the narrative hasn't attached itself to the geography yet.
Data center site selection follows a herd instinct. Northern Virginia got a head start in the 1990s because of proximity to government and early internet infrastructure, and now everyone goes there because everyone else goes there, which means the talent is there, the contractors are there, the fiber is there. Phoenix became a data center hub because it was warm enough to never deal with ice storms, power was cheap, and land was available. Momentum builds.
The Mountain West doesn't have the momentum yet. It doesn't have the marketing story. No major hyperscaler has staked a claim there at scale and written the press release that makes everyone else feel safe following.
That's exactly the window.
The Bottom Line
The AI infrastructure arms race is real, and it's going to require a massive amount of new data center capacity over the next decade. That capacity will get built somewhere. The question is whether it gets built in the places with the best marketing narrative, or the places with the best actual conditions.
The Mountain West has cold, dry air at elevation that dramatically cuts cooling costs. It has proximity to an electrical grid that's growing fast on renewable energy and still has room for new large customers. It has land that's cheap, available, and relatively easy to permit. It has low seismic risk. It has emerging tech talent in university towns. And it's connected to the rest of the country by the same fiber network everyone else uses, so there's no latency penalty for putting infrastructure there.
You don't need a rocket. You don't need a submarine. You need a site in the right zip code and the willingness to be early.
The person in mountain canyon who was heating his house in the summer. He was accidentally describing the best data center real estate in America.
Data in this piece draws on publicly available capacity figures from the Western Electricity Coordinating Council (WECC), industry efficiency benchmarks from the Uptime Institute, and the documented results of Microsoft's Project Natick experiment. The Mountain West data center thesis is independent editorial analysis.
Aaron Rose is a software engineer and technology writer at tech-reader.blog.
Catch up on the latest explainer videos, podcasts, and industry discussions below.
