There are two ways to put AI into a business, and the industry only markets one of them. The one they market is the subscription: per seat, per token, per month, climbing as you grow, billed until the heat death of the universe. It's a great business — for them.

The one they don't market is the one I'd actually want as an owner: build the thing once, own it outright, let it compound on your own data. An asset, not a bill. After a year of building exactly this for small businesses, here's how I think about the line between the two.

A subscription gets more expensive the more you use it. An asset gets more valuable the more you use it. Same tool, opposite direction — and over a few years, opposite outcomes.

The meter is the whole business model

When you call an AI API, you're renting intelligence by the word. That's fine for tinkering. But run it inside a real business and watch what happens: the better the product does, the more you use it, and the more you use it, the more you pay. Your success is their revenue line. You've built a business whose costs scale with its wins, and the meter is owned by someone with every incentive to keep it running.

That's backwards. I stopped calling APIs and started owning the model for exactly this reason. When the model is mine, the cost of running it is the cost of a rented GPU by the hour — pennies an estimate — not a per-token tax that punishes me for using my own product.

What "owning it" actually buys you

People hear "own your AI" and picture a research lab. It isn't that anymore. For a small business it means three concrete things:

  • The economics flip. Cost stops scaling with usage. You pay to build it and to run the hardware, and that's it. Use it ten times more next year and your bill barely moves.
  • The data stays yours. Your customers, your numbers, your history never get shipped off to train somebody else's product. For a lot of businesses that's not a nice-to-have, it's the whole ballgame.
  • It compounds. A model trained on your business gets sharper every time your business does something. I'm not building a model, I'm building a system that grows one — and all of that growth belongs to you, not a vendor.

The honest version of the trade-off

I'm not going to pretend ownership is free. It costs more upfront than typing a credit card into a SaaS signup. Building an asset always does — that's what makes it an asset.

So the real question is time horizon. If you need a thing for a month, rent it. If it's a commodity every business uses the same way — email, calendars, the boring infrastructure — rent that too, forever, and don't think twice. But if it's the thing that makes your business different, and you're going to run it for years, renting it is just slowly buying it over and over without ever owning it. I've written about what custom AI actually costs — and the surprise for most owners is that owning is cheaper than renting faster than they expect.

The test

Before you put any AI into your business, run it through one filter: in three years, is this an asset on my books or a line item I can never turn off?

If it's the thing that makes you different — your pricing, your knowledge, your customer history — build it once and own it. If it's plumbing everyone shares, rent it and move on. The mistake isn't subscribing. The mistake is renting the one thing that should have been yours.

That's the whole job, the way I do it: figure out which is which, then build you the asset and let you rent the rest.