A feeling is not evidence. I can tell you AI feels different from everything that came before, but you should not believe me on a feeling, and I don’t believe myself on one either. So I built an instrument. It is five questions you can ask of any information technology, and it turns “this seems bad” into something you can actually score, argue with, and check.
I call it the Diffusion-Concentration Diagnostic, which is an academic mouthful for a simple idea. For each question you give the technology a score from 1 to 5. A 1 means the power is spread out: cheap, open, in everyone’s hands. A 5 means the power is concentrated: expensive, gated, owned by a few. Here are the five.
The first four are all about the road.
D1, ownership. Who owns the capacity to build the thing? Lots of independent players, or one cartel? When AT&T bought up the independent phone operators, that became a 5.
D2, cost to enter. What does it cost to go from user to producer? Starting a blog costs pennies, so that’s a 1. Laying transatlantic cable takes a nation-state or a bank, so that’s a 5.
D3, access. Can ordinary people use it at all? A library book is a 1. A 1960s mainframe that only Ivy League researchers could touch is a 5.
D4, chokepoints. Is there a master switch? A decentralized network that routes around damage is a 1. A single wire or law that one person can flip to shut everything down is a 5.
Map two hundred years of history onto just those four and you get Wu’s cycle, exactly as advertised. The scores swing between 1 and 5 as monopolies rise and get broken and rise again. Nothing new there.
The fifth question is the one I added, and it is the entire point.
D5, the locus of generative capability. When the technology actually produces something of value, when the real creating happens, where does that act physically occur? If it happens with you, the person holding the camera, writing the code, driving the car, that’s a 1. If it happens inside infrastructure you can never own, where the road effectively drives the car for you and won’t let you steer, that’s a 5.
Now watch what D5 does across history, because this is the whole story in one line. No matter how ugly the first four questions got, D5 kept drifting toward 1. The channel got captured constantly. The capability kept ending up in our hands. Cameras shrank. Microphones got cheap. The ability to make the thing relentlessly decentralized, decade after decade. The open internet was the high-water mark: nearly all 1s, and a D5 of 1, because the power to publish and build and create lived entirely with you, out at the edges, no permission required.
Then run frontier AI through the same five questions, and the floor drops out.
D1: a 5. A tiny handful of companies own the models and the server farms. The industry calls them hyperscalers. You know the names.
D2: a 5. Training a frontier model costs billions, takes tens of thousands of specialized chips, and draws the power of a small city. The barrier to entry is about as high as any barrier in industrial history.
D4: a 5. The supply chain is a chokepoint you could not design better if you tried. The machines that make the chips come from essentially one company in the Netherlands, ASML. The chips themselves come from essentially one company in Taiwan, TSMC. Pull either thread and the whole frontier stops.
D3: a 2. This is the soft one, and it’s the trap. Using AI is cheap and effortless. You type plain English into a box. Anyone can do it. Hold that thought, because it’s the subject of the next post and it fools almost everyone.
D5: a 5. When an AI produces something, the creating happens inside a building you will never enter, on hardware you can’t buy. Not the phone in your hand. An engine in a data center behind biometric locks and lawyers.
Here is why that last score is the alarm. The only other time in the whole lineage that D5 hit 5 was the broadcast era. Every other technology, the capability stayed with the user or came back to them. AI is the second time it has ever concentrated, and the first time it’s been bolted down by physics and capital rather than by mere control of a wire.
For two centuries the arrow of D5 pointed outward, toward us. AI is the first information technology where it points hard the other way, back toward whoever owns the infrastructure. That reversal is the thing I named the Inversion, and now you can see it as a number instead of a mood.
You probably want to object. You’re holding a phone that can write a sonnet for free, and a 5 on capability sounds absurd from where you’re sitting. Good. That objection is so important, and so common, that it gets its own post.
Next: You Can’t Enter the Kitchen. Why “free to use” and “yours to own” are not the same thing, and why mixing them up is the most expensive mistake we’re making.
See the full nine-technology scores as a table: the Diffusion-Concentration Diagnostic.