Scarcity by Design
What if we could collapse the premium, the hostile debate, and the scarcity of technology simply by collapsing the fundamental belief that created them? We have come to believe that scaling compute requires scaling the site, regardless of the application. So we search for ever larger blocks of contiguous land attached to ever larger blocks of power, then marvel at how scarce they are.
But land is not particularly scarce, nor is energy in the abstract. What is scarce is large blocks of power adjacent hundreds of contiguous acres with all the other required inputs, with the contractors and capital, all intersecting in one place, on one schedule, with enough certainty that a tenant will commit.
Of course that is scarce; we designed it to be!
Then the mechanics reinforce it: the tenant wants RFS certainty before committing. Certainty requires utility commitments, equipment orders, contractor capacity, entitlements and capital. Capital wants the tenant before taking that exposure. Grow the project and every part of that loop gets harder while the number of viable sites, contractors and solutions gets smaller.
Eventually we take that increasingly rare, expensive and immovable project into an entitlement hearing. By then the community understands the leverage we created: we have narrowed our alternatives, concentrated enormous demands into one place and invested enough that walking away hurts. What should be a discussion about integrating a project into a community can quickly become a negotiation over what the community receives for allowing it.
The landowner is already saying “Don’t lowball me, I know what I’ve got.” Increasingly, the community says the same thing.
The premium is real, the hostility is real, the scarcity is real at the point of transaction. That does not mean the scarcity was inherent.
Other industries scaled without making every site progressively rarer. Walmart builds enormous facilities across small-town America by preserving site substitutability. The telephone network put dense electronic equipment, cooling, batteries and generators into quiet buildings in nearly every town, then scaled by adding nodes.
There are applications where concentration is technically justified. Large AI training systems may be one of them. But if inference becomes the much larger share of compute, we should be careful about turning the physical requirements of one application into the permanent development model for all of them.
At some point, scale became synonymous with concentration.
Maybe we did not discover scarcity, maybe we designed a model that makes only a tiny fraction of abundant resources usable, then built an entire market around paying premiums for that fraction.
If so, the opportunity may not be finding the next scarce site, it may be making fewer sites scarce.