"The earth is the Lord's, and everything in it." — Psalm 24:1. The compute, for now, is held by fewer hands.
"Anyone can build with AI now." It's the defining slogan of this moment — on conference stages, in launch posts, in the breathless coverage. The barriers are gone. The tools are in everyone's hands. A teenager with a laptop can do what once took a research lab. It is genuinely one of the most exciting promises in technology, and like the best marketing, it is true enough to be believed and incomplete enough to mislead.
So let's ask the slogan a direct question, the kind this publication has been asking all along: who actually gets to build with AI? Not who gets to use a chatbot — nearly everyone. Who gets to build something durable, defensible, and theirs on top of this technology? The answer is more layered than the slogan, and the layers tell you a great deal about how power is arranged in the AI economy.
The Three Floors of the Building
Think of the AI economy as a building with three floors, and notice who's allowed on each.
The ground floor: using. Anyone can use AI. You can type into a model and get remarkable output for free or nearly free. This floor is genuinely, wonderfully open, and the slogan is completely true here. But using is not building. Using a tool is consuming a capability someone else controls. It's powerful, and it's also the floor where you have the least leverage, because what you can do is bounded entirely by what the owner of the tool permits today and will permit tomorrow.
The middle floor: assembling. Here you build products and workflows on top of others' models — the realm of the "AI-native" startup and the operator wiring tools together. Real value gets created on this floor; most of the AI economy lives here. But as we've explored before, much of what's built here is assembled on rented foundations. You can build, but your building sits on someone else's land, reachable only as long as they keep the door open at a price you can pay. The freedom is real and the dependency is real, at the same time.
The top floor: owning the foundation. This is where the actual models get trained and the actual compute gets controlled. And this floor is not open to anyone. It is held by fewer than a dozen entities on the planet — the handful that own the chips, the data centers, the energy, and the capital it takes to train frontier systems from scratch. The slogan "anyone can build with AI" quietly means "anyone can build on the ground and middle floors." The top floor, where the deepest power and the most durable moats live, has a guest list, and you are almost certainly not on it.
The Gates Behind the Slogan
The reason the top floor stays exclusive isn't mysterious. It's gated by exactly the things this whole series has been tracing, working together.
It's gated by capital — training a frontier model costs sums only a few players and their sovereign-wealth backers can assemble. It's gated by compute concentration — the physical infrastructure is owned by a tiny number of companies, all of whom are also competitors or landlords to everyone building above them. It's gated by the "open" that closed — the research and weights that might have let others build at the foundation level got walked back behind APIs and restrictive licenses right as they became most valuable. And it's increasingly gated by regulation shaped by incumbents — compliance regimes calibrated to be survivable for trillion-dollar firms and prohibitive for everyone smaller, sold to you as safety.
Each of these, examined on its own, has a reasonable-sounding justification. Together, they form a single structure with a single effect: they keep the foundation of the AI economy in very few hands, while the slogan assures everyone else that the whole thing is wide open. The openness is real on the floors that don't threaten the concentration. The concentration is real on the floor that matters most. Both things are true, and the marketing is built to let you see only the first.
The Hypocrisy, and Its Limit
So there's the hypocrisy this series keeps circling: an industry that markets radical accessibility while engineering radical concentration. "Anyone can build" alongside "almost no one can build at the level that confers real power." It's the same move as "open" that closed, "democratized" with a seat minimum, "community" that became the product — a generous word doing public-relations work for an arrangement that is, underneath, about keeping control where it already is.
But here's where this piece turns, because cataloging the gates is only useful if it leaves you more capable rather than more resigned. The concentration is real. It is also not the whole story, and the people who benefit from it would very much like you to believe that it is — to conclude that since you can't own the top floor, you can't build anything that matters, and you should simply consume what you're given and feel grateful. That conclusion is false, and it's the most important thing to refuse.
What Is Genuinely Within Reach
You will probably never own a frontier model or a data center. You don't need to. The power available to a small operator right now, on the floors that are open, is real and historically unusual — and the gatekeepers benefit when you forget that.
You can own the parts that actually differentiate you. The rented model is a commodity; everyone has access to roughly the same one. What you build around it — your proprietary data, your specific judgment, your relationships, your fit to a particular job, the workflow only you understand — is yours, and it's frequently where the durable value lives anyway. The foundation may be concentrated, but the thing that makes your work yours doesn't have to be.
You can refuse to build everything on rented land. Knowing the dependency exists is half the protection. Keep your own data portable and your own audience reachable. Spread your dependencies so no single provider can dictate your terms. Choose tools that let you own and move your work rather than ones that trap it. You can't escape the concentration entirely, but you can decline to compound it with your own choices.
You can demand more from the people you build on. The gates aren't as permanent as they look. Every layer of this market is contestable when enough operators ask the right questions — about ownership, about portability, about who really controls the capability — and reward the providers who answer well. Tools built to serve the operator rather than capture him exist, and they grow in exactly the proportion that people learn to ask for them. Demand is not powerless. It's the only thing that has ever moved a gate.
The Cross-Examination, Concluded
This is where the series lands. The case for the AI era was stated first and loudly: open, democratized, safe, accessible, for everyone. It seemed right, the way a confident opening statement always does. Then we cross-examined it — followed the money, watched the timelines, read the structures instead of the slogans — and found, again and again, the same gap between what was said and what was built. Open that closed. Democratization with a tollbooth. Safety that doubled as a moat. Accessibility that stopped exactly where the real power began.
Naming that gap is not pessimism, and it's not anti-technology. It's the opposite. It's the refusal to let the people who hold the foundation tell you what's possible on the floors you can reach. The earth, the old line says, belongs to its maker — even if the compute, for now, is held by fewer hands than the slogans admit. Build on the floors that are open. Own what you can own. Keep your roots in your own soil. And read every generous word you're handed the way you'd read any claim from someone who profits from your belief: carefully, and to the end.
Sources: analysis of AI compute and frontier-model concentration among fewer than a dozen providers; OpenAI's own acknowledgment that scarce compute could concentrate advanced AI among the wealthy; reporting on capital requirements for frontier training, foundation-model dependency, the retreat of "open" AI behind APIs and restrictive licenses, and incumbent-shaped AI regulation (2025–2026).
