AI Is an Addictive Product on a Subscription

AI is an addictive product on a subscription. I watched a Fortune 100 company's technology organization learn what that costs.

In Orlando I met a man at a bar holding 2028 SPY puts. His bet: the economy tanks because of AI.

I told him I disagreed. Then I told him why I understood him.

I spent four years inside the technology organization of a Fortune 100 company. I watched AI arrive like a free trial and settle in like a habit.

Then the bill came due.

The company throttled usage when the cost got too high. Within weeks, work slowed and things started breaking. During one system failure, the first thing a lead engineer said was: I am out of tokens.

Not inconvenienced. Stuck.

Engineers who hit their monthly limit did not switch to the old way. There was no old way left. Removing the tool had quietly stopped being an option.

That is the part nobody prices in. Once you depend on it, it is not a productivity tool. It is a utility. And a utility with one supplier sets the terms.

I am not judging from a distance. Even I have a hard time not using it. That is the point.

The same math is running in public now. Uber cut about 3,300 jobs yesterday, roughly ten percent, while the same announcement committed more than ten billion dollars to autonomous vehicles. One employee posted: apparently, like everyone else, we can fund AI.

Microsoft cut 4,800 people in July while reportedly planning one hundred ninety billion dollars for AI infrastructure and data centers this year.

Follow the money, because money is finite. OpenAI is reportedly targeting thirty billion dollars in revenue this year. That revenue is not printed. It is other companies' new costs: subscriptions, tokens, compute. And when a cost that size lands on corporate budgets, the largest line item anyone controls is people.

I cannot prove any single layoff paid an AI bill. I can read the announcements, where the cuts and the AI commitments keep arriving in the same paragraph.

And the company collecting it is heading toward the public markets in a trillion-dollar conversation: thirty billion coming in, fourteen billion projected lost, backed by the same few giants that supply and host it. When that circle goes public, the dependency stops being a corporate problem. It lands in index funds. In retirement accounts.

I do not know if the man at the bar is right about the crash. That is his bet.

Mine is smaller, and I can build it: never depend on one vendor's meter.

That is what the dream file is for. A written contract for the work: what the system should become, what done means, and it does not care which model reads it. Models change. Prices change. The contract stays.

He hedges with puts. I hedge with portability.

Day 3 of documenting what I learn building with AI.

What breaks in your business if one vendor changes one number?

Farewell until tomorrow.

Come build with me: xiv@marcelozapata.dev

HE HEDGES WITH PUTS. I HEDGE WITH PORTABILITY.