Intelligence Is the New Currency — and the Exchange Is Infrastructure

Tomorrow, Explained · Issue #1

You have heard the sentence a hundred times this year: artificial intelligence is going to change everything.

It is true, and it is useless. It tells you nothing you can act on, hold, or plan around. It is the kind of sentence that makes a technology feel like weather, something that simply happens to you.

So let me offer a different one, because it is the sentence this entire publication is built on:

The exchange is infrastructure: the compute, the energy, the networks, and the physical stack that turn a clever model into something the world can actually use. That layer is where the next decade is being decided. It is also the layer almost no one is looking at, because it is buried under two kinds of noise.

On one side, hype. AI will cure every disease, write every email, replace every job by Thursday. On the other, jargon: a wall of acronyms and benchmarks so dense that a smart, busy person quietly concludes the whole thing is above their pay grade and stops trying to understand it.

Neither serves you. The hype tells you everything is changing without telling you what. The jargon tells you what in a language designed to keep you out.

Between them sits the thing that actually matters, and it is more legible than either camp lets on. You do not need a PhD to understand it. You need someone to translate, to point at the currents under the surface waves and say, plainly, here is what is moving, here is why, and here is what it means for you.

That is the job.

For most of computing history, intelligence — the ability to reason over a problem — was scarce and expensive. You hired it, in the form of people, or you did without.

That is changing. The cost of a unit of machine reasoning is falling fast, and as it falls, intelligence starts to behave less like a rare skill and more like a resource you pour into a process, the way you would electricity, or credit, or bandwidth. Something you meter, budget, and build on top of.

But here is the catch that the "AI changes everything" crowd keeps missing. A currency is only as good as the exchange that clears it. Money is worthless on a desert island. Intelligence is worthless without the infrastructure to generate it, move it, and put it to work. And that infrastructure is neither cheap nor abundant nor evenly distributed.

Which is why the real questions of this decade are not the ones dominating the coverage.

The question is not which model is smartest. At the frontier, the gap between the best models narrows by the quarter, and whatever gap remains, you can rent. The question is who owns the exchange — the compute and the power and the pipes underneath.

Let me make the exchange concrete, because vagueness is how it hides.

Compute is simply raw processing power: the specialised chips, and the electricity to run them, that do the actual work of AI. It sounds like a spec on a datasheet. It is really a question of who holds the means of thinking, because compute is physical, finite, and concentrated in remarkably few hands.

Inference is the model doing its job. Every time you ask it something, that is inference. Training a model is the giant bill you pay up front, and pay again with every major update. Inference is the bill that never stops: it comes due every single time anyone, anywhere, uses the model. And it is becoming the expensive part.

The stack is everything underneath the friendly chatbot window: the chips, the data centres, the energy contracts, the software that ties it together. When people say the competition has "moved down the stack," they mean the advantage is no longer in the model you see. It is in the layers you don't.

Learn those three words and the headlines rearrange themselves. The AI story stops being a horse race between chatbots and becomes what it actually is: a contest over infrastructure.

Here is one figure to carry with you. Data centres — the physical homes of all this intelligence — consumed roughly 415 terawatt-hours of electricity in 2024, about 1.5% of the world's total. On current trends that figure roughly doubles by 2030, to around 945 terawatt-hours, nearly 3% of all the electricity on Earth (IEA, Energy and AI).

The bottleneck on artificial intelligence is turning out to be the least futuristic thing imaginable: the electrical grid. You can design a brilliant model in months. You can wait years for a power connection to run it.

The frontier of AI is not a laboratory. It is a substation.

That is what it means to say the exchange is infrastructure. And it is why so much of what gets sold as a story about software is really a story about steel, copper, water, and power: about physical things, in physical places, owned by specific people.

If you lead a business, AI is not a tennis match between models where you pick a favourite. It is a long game of infrastructure strategy: what you build on, what you depend on, and what you can move when the terms change.

If you make policy, compute is becoming as strategic as oil. You either help shape access to it, or you depend on the people who do.

And if you are simply trying to understand the world, the value of intelligence over the coming years will be settled less by clever algorithms than by who controls the infrastructure beneath them. That is a question about power, in every sense of the word.

Most people will keep watching the surface waves of AI: the launches, the demos, the benchmark records. Here, we will chart the currents underneath. The ones that decide who wins, who loses, and what kind of society we build on top.

Because the next five years will not be won by the best model.

They will be won by the best infrastructure — and by the people who understood that early enough to do something about it.

That is the lens this publication will keep handing you, one issue at a time: not which model wins, but who owns the exchange it clears through. If that is the version of the future you want to understand — the one actually taking shape, not the one shouted at you — you are in the right place.

Next edition — Kilowatts Are the New Borders: why a data centre is not a building but a claim on a substation, a river, and a permitting queue, and how energy quietly became the first border of the AI age.

If this made the picture clearer, forward it to one person who keeps asking you what's really going on with AI. That is how this grows.