Can Real Estate Kill AI?
One of the most interesting things about AI is its deep dependence on the physical world, even though we typically imagine it as an ethereal, digital brain living in the cloud. We’ve talked about how power is the single biggest bottleneck to AI’s growth. But there’s another factor that’s becoming equally critical: real estate.
The Physical Side of AI
AI doesn’t just exist on the internet—it lives in machines. These machines are packed into massive data centers filled with servers that perform millions of computations every second so you can get an instant answer from a large language model.
When I had my first startup, even though we used AWS for much of our work, we still leased our own rack space—our own little server farm in Silicon Valley. I remember visiting it on an 80F day in shorts and a t-shirt, then walking into a freezing-cold server room where our operations guy always wore a winter jacket.
Why? Because if a server’s environment gets above about 95F it starts to fail. Too cold—below 50F —and components can malfunction. The sweet spot is somewhere between 64F and 70F.
Maintaining that balance costs a lot in cooling, energy, and space. This brings us to the next part of the story: where all this infrastructure happens.
The Hidden Land-Grab of the AI Age
Big tech companies have server farms all over the country, often called co-locations or colo sites. For example, some of the most famous ones are located in the Nevada desert.
Why there? Because the real estate was dirt cheap, the air was dry (which aids in air-cooling), and the state offered incredible subsidies. Companies leveraged air-cooling technology that took advantage of Nevada’s low humidity to keep massive server farms cool without needing expensive, water-based systems.
But not every place is that simple. Recently, several towns in the U.S. Midwest rejected new Google data centers because of water-usage concerns. Google planned to use water-based cooling—imagine a radiator system, but instead of hot water heating your apartment, it’s cold water cooling racks of GPUs. Town officials pushed back, noting that such facilities could consume a massive share of the community’s water supply.
Meanwhile, regions like Pennsylvania and New Jersey have become prime targets for smaller data centers because they are close to major metros like New York, D.C., and Philadelphia. They offer a critical balance: cheaper real estate than the big coastal metros and lower latency for network connections. However, they lack the ideal climate for passive cooling, which means more cost and power are needed to keep things stable.
And that is the paradox: AI wants to grow infinitely, but it still needs finite, real-world space.
When Zoning Meets Zeros and Ones
Cities and counties are now waking up to the true costs of being an “AI hub.” Data centers consume huge amounts of electricity and water. They drive up land prices. They require new substations, power lines, and roads. And most run 24/7 with very few employees.
Take Northern Virginia—the densest cluster of data centers in the world. Utilities there have warned that new facilities are already pushing electrical grids to their limits. Phoenix and Dallas are battling over water rights. In parts of Oregon and Iowa, farmland is being converted into server-compound real estate. The result? The same land that once grew crops or housed families is now being turned into compute infrastructure for large language models.
The irony is striking: we’re building machines that can think faster than any human, but they are constrained by the very basics of Earth: land, power, and water.
Each new AI model demands exponentially more compute. Compute needs chips, chips need power, and power demands space. The idea of “infinite digital intelligence” runs straight into the wall of finite physical resources. That’s why the new AI boom looks a lot like the Industrial Revolution—only instead of steel mills, we have data mills. And the land they occupy is becoming one of the most strategic assets in the AI economy.
The Statistics You Should Care About
Here are some real metrics to ground the discussion:
Globally, data-centres consumed an estimated ~460 terawatt-hours (TWh) in 2022 — roughly 1.4%–1.7% of global electricity usage. Canada Energy Regulator+2IEA+2
In the U.S. in 2023, data-centres used roughly 176 TWh, or about 4.4% of total U.S. electricity consumption. The Department of Energy’s Energy.gov+2TechTarget+2
The global power-demand for data-centres is forecast to grow ~50% by 2027 and up to ~165% by 2030 (relative to 2023 levels). Goldman Sachs+1
Electricity operating costs make up about 20% of total cost for data-centre business models — meaning owners are highly sensitive to power, and willing to pay premium rates. McKinsey & Company
These statistics show the scale of what we’re talking about—huge power draws, rapid growth, and massive infrastructure implications.
The Coming Real-Estate War
So, what does the future look like? Essentially: a real-estate war, with the biggest companies in the world fighting for the most optimal locations to build massive co-locations and server farms.
If a company finds a place that’s cold, stable, with access to clean power and cheap land, that location is going to be gold.
Examples:
A place like Greenland—cold climate, abundant hydropower potential, and one of the few remaining “unused” terrains.
The mountains of Pennsylvania—cooler air, lower moisture, decent real-estate pricing, and proximity to major Northeast metro markets.
Underground facilities? Imagine massive data farms built beneath the surface for temperature stability, security, and latency control. This is feasible and increasingly considered.
Also, towns in the middle of America (or Canada, or other global regions) that have been hit by economic decline could see a resurgence if they can attract “AI-farm” real estate. A facility that employs a few hundred directly might enable thousands of indirect jobs (construction, operations, local services). Real estate becomes the battlefield for AI infrastructure.
Why Real Estate Can Constrain AI
“Kill” might be dramatic, but the point is this: the bottleneck for the next leap in AI won’t just be better models or faster GPUs. It will be where we physically house those GPUs and how we power and cool them without crippling the planet or the grid.
If large tech firms cannot secure:
Land
Cooling mechanisms (air vs. water vs. immersion)
Access to power (clean or subsidized)
Proximity to networks (latency matters)
Favorable local policy (zoning, water rights)
...their ability to scale AI will be severely constrained.
In other words: Great algorithms + great hardware + bad real estate or bad infrastructure = stalled innovation.
The Future of Real Estate in the AI Era
Expect “AI belts” to emerge across middle America and other geographies: cheaper land, abundant power (especially renewables or next-gen nuclear), and moderate climates.
Meanwhile, “AI offices” in coastal metros will interface with remote compute clusters. The expensive real estate near urban markets becomes less about hosting thousands of racks, and more about access, design, and user experience.
Real-estate investors are already shifting: tracts of land near transmission lines, hydro dams, and fiber hubs are becoming highly strategic.
Real-estate zoning, water-rights, local community push-back, and cooling technology adoption will become key strategic decisions for tech firms—not just chip-design, software architecture, or cloud strategy.
Final Thoughts
Can real estate kill AI? Not in the dramatic “end of intelligence” sense, but it can absolutely limit it. The next wave of AI innovation will depend less on just models or chips, and more on zoning laws, water permits, land availability, and cooling strategy.
In the race for AI dominance, the biggest companies may find their most strategic resource isn’t data scientists or GPUs—it’s real estate.