The ground is moving so quickly in a town west of Abilene, Texas, that you don’t really believe it until you stand next to it. There are excavations, steel frames, and transmission lines that go off into the distance. A company called Crusoe is building what might be the densest collection of computing power ever put together in one place. The company started out by burning waste gas to mine bitcoin. There’s no real sense of what a data center is here. The people who are building it have started calling it a factory, which doesn’t seem like a joke. It takes in energy. It comes out that intelligence is measured in tokens.
It used to be that word, “tokens,” was only used by scientists and engineers. It is now part of business. You’ve probably seen what these tokens are increasingly being used for if you’ve recently bought something online, agreed to buy something suggested to you, stayed on a page longer than you meant to, or clicked on an ad that seemed almost too good to be true. There won’t be a convincing tech boom. Bills are already being sent.
About 30 trillion tokens are processed every day by Fireworks AI, a San Francisco-based inference platform that most people have never heard of. Companies like Uber, Shopify, and Notion are among its clients. The fact that people ask chatbots questions doesn’t cause that number—30 trillion—to appear. It takes place because software calls itself over and over again. One click from the user can set off a chain of model calls that read the context, make assumptions about the behavior, get data, and make a response that is meant to keep the conversation going. Agents no longer just answer the phone. They do something. And more and more, they work for companies that want to move goods.
This might sound vague at first until you think about what it means in real life. There have been recommendation engines for a long time, but the ones being made now are different, if not always in the same way. These systems don’t just tell you what to do next. They keep track of patterns of hesitation, change messages in real time, and lead users through flows that were designed by optimization rather than intuition. It’s never been clear where the line is between helpful personalization and engineered nudging. It’s getting harder and harder to find.
As I watch this happen, I get the impression that the infrastructure buildout in places like Abilene isn’t just about how fast things can be processed or how much space they take up. It has to do with reach. The price per token, which insiders now call the sector’s “effective interest rate,” has been falling quickly. Whenever it does, products that didn’t seem possible before suddenly become viable. It’s now possible to use new AI-powered tools to target, keep, and convert customers on a large scale that wasn’t possible even eighteen months ago. AI isn’t just getting faster thanks to companies like d-Matrix that are making chips that make real-time inference much cheaper. They are lowering the cost of persuasion.
Not only is this era different from the behavioral advertising wave of the 2010s, but it’s also coordinated. The systems that are being put in place now don’t work separately. They use the same context in different products, sessions, and platforms. Not only are they being used to show ads, but they are also being built into how people make decisions. Agents work for Shopify merchants. More and more, customer service flows are controlled by agents. More and more, the moment you decide to buy is influenced by a model that guesses what will likely make you go through with the deal.
Nothing about this is really hidden. They don’t say it out loud, though. There is still some doubt about whether regulators or most users fully understand how much of the commercial internet has already been rewired. The factories are up and running. The tokens are coming in. The bills are real. What’s less clear is whether the people who are being inferred about have any real control over how it’s used against them. It’s possible that that question, more than any earnings report or valuation round, will make or break this boom.
