Hey friends,

One month last year my AI bill came to around fifteen thousand dollars.

Not a typo. APIs, consumption, systems running in the background, agents doing work while I slept. I was building heavily that month and I burned through it without really watching.

When I mentioned this on the podcast, Davina had this image of me sitting in front of a small fire made of money. Which is close enough to accurate that I did not argue.

But the number is not the point. The point is what it tells you about what the rest of us are currently paying.

Here is the piece most people have not sat with properly.

The usage you get from a thirty dollar Claude or ChatGPT subscription would typically cost the provider a hundred dollars or more to deliver. Sometimes a lot more. They are absorbing the difference, deliberately, funded by the enormous amounts of capital they have raised.

They are not doing this out of generosity. They want as many people as possible using AI, forming habits, building things they cannot walk away from. That is a completely rational strategy and it is working.

But it means the price on your invoice is not the cost of the thing. It is a marketing number.

Meanwhile the underlying cost of tokens has not really come down. There has not been much progress on that part of the problem. And a token is the actual unit of the whole thing, roughly a word, sometimes a fragment of one. Every message you send and every response you get back is measured in them. Whether or not you are paying per token, somebody is. There is real compute happening, on real hardware, drawing real power.

We also have not solved the energy question underneath all of it. Somebody needs to, and I assume good people are working on it. But it is not solved today.

I think it is entirely possible that in a year, the twenty dollar subscription is a five hundred dollar subscription. I am not predicting that with confidence, and nobody has announced anything like it. But if you are planning a business around today's pricing holding, you are planning around someone else's fundraising round.

Now, and this is important, that is not a reason to slow down. It is the opposite.

Right now you have access to genuinely powerful capability at a price somebody else is largely paying for. That is a strange and temporary situation. The correct response to a temporary discount is not caution. It is to use the window for what it is actually good for.

This is the year to experiment, learn, and fail cheaply.

Burn tokens. Try the thing that probably will not work. Build the internal tool nobody asked for. Push a system until it breaks so you understand where it breaks. All of that costs money you would not have been able to spend two years ago and might not be able to spend two years from now.

What you are buying with that spend is not output. It is knowledge of where the waste is.

Because there is a lot of waste right now, and most of it is invisible. Ordinary conversation is cheap. I barely think about it. But when you are building applications, designing systems, running agents against large sets of files, you are not burning thousands of tokens. You are burning millions. I know this because I have watched it happen on my own account.

Nobody has properly worked out yet which of that consumption produces value and which is just motion. That is the actual thing to solve before the price corrects, and you cannot solve it theoretically. You have to spend the money once to find out.

I built an app over a weekend and two nights after it.

Before I started, I asked my team for an estimate on the same build. I have a genuinely skilled team, they are not slow, and they do not pad numbers. Three and a half months.

So yes, I spent real money that month. I also compressed three and a half months of skilled engineering time into about four days.

Put those two facts next to each other and the fifteen thousand looks different. Not because the spending was clever, but because time has a price too, and we have all been quietly pretending it does not.

The dumb monkey move is treating the subscription price as the real price and building your operating model on top of it.

I see businesses rolling out AI across a hundred seats, celebrating how cheap it is, with nobody tracking what any of it produces. When the price moves, and it will move in some direction, they will have no basis for deciding what to keep. They will cut the wrong things, because they never knew which things were working.

The opposite mistake is just as bad. Refusing to spend anything until the pricing settles, then arriving late with no institutional knowledge and paying full freight to learn what everyone else learned during the cheap years.

Pick one task your team does repeatedly. A report, a proposal draft, a first-pass analysis, whatever comes up weekly.

Run it through AI properly, and write down two numbers: roughly what it cost in tokens or credits, and roughly how many hours it saved.

You now have one real data point about your own business rather than a general feeling about AI. Do that four or five times and you have something close to an actual economic model, built while the inputs are still discounted.

The bill is coming. Not as a catastrophe, but as a correction. These companies have extraordinary cost bases and valuations that assume the money eventually flows the other way. It will.

None of that should stop you from using AI now, and I want to be clear about that, because "there's a bubble" has become a convenient excuse for doing nothing.

The businesses that come out of this well will not be the ones that predicted the pricing. They will be the ones who used the discounted years to find out precisely where AI creates value in their own operation, so that when the number changes, they already know what is worth paying for.

Everyone else will be making that decision under pressure, with no data, in a hurry.

See you next week,

— Aamir

🎧 Listen to the Podcast Episode on: Spotify | Apple Podcasts | YouTube

📘 Book: The CEO Who Mocked AI (Until It Made Him Millions) by Aamir Qutub

🚀 Dumb Monkey AI Academy

📱 Dumb Monkey AI Academy App: Apple | Android