If you’re looking for the future of AI, don’t look at San Francisco. Look at Abilene.
Abilene, Texas — population 125,000, more churches than startups, home to an active Air Force base and a few cattle auctions — is quietly becoming a node in the global AI substrate. And not because it has top-tier researchers or a founder scene. It doesn’t. That’s the point.
Abilene isn’t building apps. It’s building the physical substrate of cognition.
It has something more valuable than talent or capital: land, power, fiber, and silence.
Welcome to the Machine Zone
On the outskirts of Abilene, between wind farms and rail lines, Crusoe Energy is building one of the largest AI data center campuses in North America. At full buildout, it will house up to 1.2 gigawatts of capacity — enough to run 400,000 Nvidia GB200 GPUs.
Oracle alone has committed to buying $40 billion worth of Nvidia compute. Much of it will spin in this dusty corner of West Texas. This is not a data center in the traditional sense. It’s a compute citadel. Call it a GPU furnace engineered to soak up electrons at planetary scale. And it’s not alone. The Stargate Project, a $500 billion hyperscale AI infrastructure initiative led by OpenAI, Oracle, and SoftBank, is laying down a new kind of grid across the U.S., with Abilene as one of its anchor nodes.
The rationale is simple: to train and serve intelligence at global scale, you don’t just need GPUs. You need geography.
The Inversion: From Code to Concrete
The fantasy of AI still gets marketed as weightless: all in the cloud, ethereal, placeless. But that fantasy is collapsing under the weight of physical constraint.
Here’s what AI actually needs:
Proximity to gigawatt-scale electricity
Low-latency access to fiber backbones
Permissive land use regimes that won’t tie you up in 3 years of NEPA litigation
Cooling, water, land, and footprint
Abilene checks all those boxes, and almost nothing else.
What used to be bugs are now features. It’s flat, cheap, underbuilt, and politically friendly to massive industrial infrastructure. In an era where AI is becoming the dominant load on the grid, that makes Abilene sacred land.
The Logic of Locational Arbitrage
Why Abilene?
Because Austin is full. Phoenix is overheating. Northern Virginia is congested. New York and California are allergic to permitting.
The cloud is becoming real estate. And data center developers aren’t chasing talent anymore. They’re chasing:
Tax abatements
Right-of-way fiber corridors
Megawatts on demand
Cities with planning commissions that say yes
Abilene offers all of that and none of the noise. In the new AI economy, obscurity is alpha.
AI as National Infrastructure, Not a Startup
This is the deeper shift most people miss.
AI is no longer a product category. It’s infrastructure: critical, physical, and sovereign.
It’s not “ChatGPT for X.” It’s not an app. It’s a utility, like electricity or water. And it demands the same foundation: substations, transformers, turbines, land rights, cooling systems, and national-scale energy policy.
The companies that get this, including OpenAI, Oracle, and Microsoft, are racing to build vertically integrated AI infrastructure, from the model weights down to the turbine. That’s what Stargate is. That’s what Abilene is.
You cannot train GPT-6 in a WeWork. You need a battlefield of silicon, steel, and surplus electrons.
The Strategic Implications
Here’s the real thesis: The map of AI is being redrawn, not by founders, but by land use lawyers, grid operators, and infrastructure billionaires.
The next decade of AI won’t be defined by who builds the best agent UI. It’ll be defined by who can:
Lock up power at cost
Deploy compute in bulk
Serve inference loops at scale
Do it all without triggering regulatory blowback
This world doesn’t reward flashy design or Bay Area real estate. It rewards low-friction land, permissive planning commissions, and cities like Abilene.
The Moral: It’s All Physical Now
If your mental model of AI still treats it like software, you’re going to miss the whole game. This isn’t about models anymore. It’s about megawatts. The infrastructure layer, not the app layer, is where the choke points are. It’s where the value accumulates. It’s where strategic control happens.
The future of intelligence will not be distributed. It will be substationed: rooted in electrical transformers, turbines, and quiet deals no one hears about until it’s too late.
If This Resonates
If you’re new here, welcome. I write daily about the physical substrate of AI: power, land, geopolitics, model scaling limits, and national compute strategy.
Some posts are free. Others are paid. But all follow the same principle: Cut through the noise. Look at who controls the power.
If you liked this:
→ Read Why OpenAI Will Eat all the GPT Wrappers
→ Go paid for my deep dive on ERCOT, the Texas grid quietly becoming one of the most important AI regulators in the world
→ Or send this post to someone still betting on wrappers, and tell them to look harder at substations
👀 And tomorrow’s paid post goes deeper into Abilene:
How Crusoe secured power, land, and fiber
The permitting arbitrage behind Stargate's site selection
Why Abilene could become the next Ashburn, VA, but optimized for GPUs instead of Gmail
You’ve just seen the 30,000-foot view. Tomorrow: we hit the dirt.
Subscribe to get it:
If you want to connect with me directly, you can:
Meta just contracted in Illinois to re-open and operate a nuclear reactor. https://www.cnbc.com/2025/06/03/meta-signs-nuclear-power-deal-with-constellation-energy-.html Greenies wanted to close it
A retired Air Force base? Dyess is quite active.