I use these tools every day, and some of what makes them cheap troubles me. Here’s how I’m thinking about four costs I’d rather not see, and what I plan to do about them.

In 2010, I wrote Seven Languages in Seven Weeks, and here is a confession I’ve made before: I wrote that book because I was scared.

The ground was moving under me. New languages were arriving faster than anyone could track, and the people I respected most were walking off the terrain I’d built a career on. In my circle, my heroes were moving to Clojure, Ruby, Scala, and Haskell. They were chasing functional ideas, fresh approaches to concurrency, and entirely different ways of thinking about a program. I didn’t understand why the smartest people I knew were leaving, and that failure to understand frightened me. So I learned seven of those languages in seven weeks and I wrote down what I found as I found it. I decided to bring my readers along, not because I had the answers, but because I knew that others were curious, interested, and even afraid much as I was.

Fifteen years later, the same feeling is back, and this time the thing on the move is AI. It has arrived faster than any technology I’ve watched. ChatGPT reached a hundred million users in two months, quicker than any consumer product before it, and the companies building these models are now pouring hundreds of billions of dollars a year into the data centers that run them. The consequences, good and bad, are landing just as fast, and nobody has finished counting them. So I’m going to do what I did in 2010.

I use these tools every day. Last Sunday an agent planned my family’s menus around a heart-healthy diet, a grocery budget, and the two nights I knew I’d get home late, and it missed nothing. My woodworking is more productive than it has been in years, because when I get stuck on a joint or a finish, I have a patient shop assistant who has read everything. And as a heart patient, I use these models to sanity-check my doctors, not to replace them, but to walk into the appointment knowing which questions matter.

That last one is not a party trick. It is my life, and the tools make it better.

At work it’s the same story from a different angle. The languages and the frameworks are improving quickly. I can train a team without burning the first week on vocabulary, and I can write a sharper proposal, built from the notes of a real conversation, in an afternoon that used to take a week.

So when I turn critical, understand that I’m not doing it from the bleachers. My hands are in these tools all day, and I like most of what I find there.

But I can’t read the price.

Look at what a token costs, and you get a number that means almost nothing. It’s subsidized, and the figure on the screen sits nowhere near what the thing costs to produce. Half the market runs on free tiers that hide the cost entirely, and even the paying customer isn’t seeing the real number, because the whole industry is selling below cost to win you. Before we get anywhere near ethics, the dollar price is already a magic trick.

That price was never the whole bill anyway. There are costs that never appear on an invoice: the electricity and water these systems consume, the human work that trained the models, the jobs displaced along the way, and the money and influence that all of it concentrates in very few hands. None of that is printed on the label, because there is no label. Every AI pitch is the happy part of a drug commercial, the kitchen and the dog in golden light, and this series is the fast, calm voice at the end that lists what it may cause.

I want to itemize four of those costs.

The first is the data center, which sends a physical bill in power, water, noise, and a town of strangers that pays for all of it while the profit is booked somewhere else. The second is intellectual property, the books and code and art and music that trained these models, made by people who were never asked and never paid. The third is work, and what becomes of the junior on-ramp and the human parts of the craft once the machine does more of the producing. The fourth is wealth and power, which comes down to who ends up owning the upside and who governs something this consequential.

I’m going to skip some things. I won’t write about autonomous weapons, the security nightmares, or whether a superintelligence wipes us all out, not because those questions don’t matter, but because they’re already covered everywhere, loudly, by people who know them better than I do. The boring bill, the one already sitting on your receipt, is the part nobody itemizes, and that is the part I can help with.

I owe you fair warning. This guide is going to make the price more confusing, not less, and I won’t hand you a clean number at the end, because there isn’t one. Some of these costs you can act on as a buyer. Others no shopping list can fix, and pretending otherwise is the comfortable lie I’m trying not to tell. The point isn’t a tidy answer. The point is to see the mess clearly, which is the one thing the golden-light commercial will never help you do.

So here is the plan. Over the next few pieces I’ll walk through those four costs one at a time, tell you where I’m complicit in each, and end with what I actually plan to do about it. This is the same impulse that drove the book in 2010. I don’t have it figured out, and I’d rather work it out in front of you than pretend that I already have.

Come argue with me. That’s how I learn. And if these values sound like yours, Groxio is at grox.io.

Sources

• ChatGPT reached 100 million monthly active users in January 2023, roughly two months after its late-November 2022 launch, the fastest ramp of any consumer app on record: Time (citing Similarweb data and a UBS analyst note)
• The four largest hyperscalers (Amazon, Microsoft, Google, Meta) spent roughly $260 billion in combined capital expenditure in 2025, up about 52% year over year, the bulk of it AI data center buildout, per their quarterly earnings: Shield Operations summary

— Bruce