🛑 The End of Unlimited AI

I posted the following graphic to CTRL+ALT+BUILD this morning.

For the first time ever I passed 1 billion tokens in a 30 day window on a single service. 

Between my OpenRouter, Codex, OpenAI, mini-apps, ventures, Cursor subscription and all my client accounts, I'm around 1.5billion+ or so.

But this was the first time I pushed a billy on my own personal Claude account in a single month. 

If i would have paid for those tokens? over a grand. 

Total spend across all the above? ~$200. 

And [drum roll], the out of pocket cost of that 1.2b tokens? $100 for my monthly Claude plan.

Right now, you are using the most powerful AI in history at a price someone else is paying for.

ONE TENTH of what they would have charged, in my case.

I know.
That's surprising, right?

In a capitalistic market, someone else is footing the bill for millions & millions of users?

The truth is that Frontier Labs subsidizes what we pay. 

We are in the preview window, and previews do not last forever. 

I said it bluntly on a call last week: the days of subsidized, flat-rate AI are nearly over.

We need to milk this stuff for everything we can. 

And the time is now.

Key Facts

• 📉 The GitHub Shift. On June 1, GitHub Copilot moved to usage-based billing. One developer's typical usage that cost $28 flat would have been around $700 metered. That is a 25x difference.
• 🛑 The Anthropic Cutoff. In April, Anthropic blocked Claude Pro and Max subscribers from using their flat-rate plans with third-party agent frameworks, pushing them to pay-as-you-go. Some users saw 10x to 50x cost jumps overnight (I was one).
• 💸 The $14,000 Subsidy. A fully utilized $200-a-month ChatGPT Pro plan can reportedly generate up to $14,000 worth of API-equivalent compute if you play it correctly (similar to my case above).

Why I think the meter is about to change…

When I got access to the newest Frontier models over the last few months, two things hit me at once.

1. The power.

I fed some of my codebases to Claude Fable 5 and it found security issues I did not even know existed. 

I sat back in my chair and thought wow, this is incredible.

2. The cost.

These flagship models are expensive AF.

So expensive that reprocessing one client's dataset on the top tier would run hundreds (to possibly thousands) of dollars.

The labs cannot eat that forever.

My gut feeling?

When Anthropic IPOs and all their numbers go public, they could easily say "here is a $20-a-month plan, and everything else is enterprise or pay per use only." 

And there is nothing anybody can do about it.

In the next few months, we will see a lot less of the "pay a couple hundred bucks a month and get tons of usage" plans. 

It is very likely going to move to per-usage. 

And that can will rack up fast.

It costs them a lot to actually process everything and we're already seeing the cracks in the flat-rate model.

ℹ️ What This Means for You

When pricing flips from flat to metered, the businesses that win will be the ones who already did the heavy, expensive, one-time work while the compute was cheap.

The expensive part of AI is not the monthly subscription. 

It is the setup: getting everything your company knows into one place and turning it into something your whole team can use. 

Do that now, while a frontier model will chew through your entire operation for the price of a nice dinner, and you have locked in an asset that keeps paying off after the meter turns on.

👉 If you run any kind of business, sit with these for a minute:

→ Where are you holding back on loading your company data into AI because you think you have time?

→ If your flat-rate AI subscription flipped to a metered API bill tomorrow, which workflows would suddenly be too expensive to run?

✍️ Use This Prompt This Week

Here is the fastest way I know to get on top of this.

Open the AI tool you already pay for, paste this in, and watch what it tells you. 

It turns the AI into a CFO that audits your own usage and puts a real dollar figure on the thing you have been getting for free.

"Act like a sharp CFO reviewing our AI spend. Based on what you already know about my business and how a company like mine typically uses AI, list my five most likely high-volume AI workflows yourself, then confirm them with me. For each one, do five things. First, estimate the rough monthly token volume. Second, calculate what it would cost at current metered API rates on a frontier model like Claude Opus 4.8 or GPT-5.6. Third, compare that to the flat monthly subscription I pay today. Fourth, flag which workflows would become too expensive to run if pricing flips from flat to per-usage. Fifth, tell me the one-time setup work I should do right now, while compute is cheap, to protect my highest-value workflows. If you are missing anything you need, fill in your best assumption and label it as an assumption."

Picture the output for a second. 

A clean little table with your real workflows down the left side, your $20 or $200 flat sub in one column, and the true metered cost sitting right next to it. 

For most people, that second number is five or ten times bigger than what they pay, and a few of their favorite workflows light up red as too expensive to keep running once the meter turns on.

That gap between the two columns is the subsidy. 

Right now, someone else is paying it for you.

The whole point of this exercise is to see, in plain numbers, exactly how much value you are pulling out of a window that will not stay open.

💭 A simple way to think about ROI here: the work that is expensive to set up once, and cheap to run forever, is the work you want done while the meter is still off. 

1. Load your knowledge in.
2. Map your workflows.
3. Train your team on a system that remembers how you operate.

You pay the setup cost one time at today's prices, and that asset keeps paying off long after per-token rates climb. 

The teams that wait will pay premium rates to do the same foundational work, slower, and behind everyone who moved early.

🧠 The Company Brain Build

When you run that prompt and stare at the red rows, the natural next question is what to actually do about it. 

And the honest answer is that the highest-leverage move is also the one most people keep putting off, which is doing the heavy setup once, now, before it gets expensive.

That's the whole reason I build Company Brains like I've done for myself. 

I take everything your company knows, how you operate, what you sell, your workflows and documents, and I load it into one unified AI layer your whole team runs on. 

You do not need another generic agent from Instagram.

You need your own, set up correctly, while the economics are still in your favor.

The window to build cheap is open right now. I would use it.

I only take a limited number of these builds each month, and the price will go up soon. 

If you want me to build your Company Brain before the meter changes here's the link while I have a few more spots still open.