How do I buy AI?
The market looks like a flood, but there are really only four ways to buy AI. Four that cover essentially everything. Here they are, then a short path to picking which one is yours — before you spend a cent.
See the four ways about 3 minutes
These four ways cover essentially everything a business buys. Read them, and see which one sounds like you.
The all-in-one platform
A flat monthly fee, no token math — AI built into a product you use. Right for most owners.
The pay-per-use API
You build something and pay per chunk of text. Right for a business with a developer.
The cheap, fast tier
High volume, low price, but you must know how to use an API. For cost-sensitive builders.
Run it yourself
Sovereign and private — AI on your own hardware. For when data can't leave your building.
Which of these is you?
Read each card once. The one that describes your situation is your box — you don't need the detail yet, just the fit. If none fits cleanly, step 2 walks you through the decision.
Done when: you can say which one sounds most like your situation.
Pick your box about 10 minutes
Two questions narrow you to one box. Answer them in order.
How to pick
Question 1 — do you want a ready-made tool, or AI built into something of your own?
- Ready-made tool (owned or bought new): you're in all-in-one. Done — skip to step 3.
- Building into your own thing: keep going.
Question 2 — if you're building, how much volume and how cost-sensitive are you?
- Moderate volume, want a name-brand, reliable API: pay-per-use API.
- High volume, cost matters, you can engineer around quirks: cheap & fast tier.
And one rule overrides all of it: if your data can't leave your building, the cloud options are out — you run it yourself, whatever the other answers.
Done when: you've picked one box. Open its card to read the detail.
Check it before you commit about 20 minutes
For the box you picked, verify what it costs and what reliability to expect — against the provider's current page, because prices move.
How to check it
- Open your box's detail page from the cards above and read what's there.
- Look at the price shape, not the sticker. Flat monthly? Per token? Hardware + upkeep? Match it to your volume.
- Check the real number at your size. All-in-one: how many seats? API or fast tier: what at your volume? Run it yourself: what's the hardware?
- Set your reliability expectation. Mature vendor with occasional outages? Or your own hardware, your problem? Budget for it either way.
- Confirm against the provider's current page. Don't trust a price you read anywhere else.
Done when: you know roughly what it'll cost and what reliability to expect.
Commit small ongoing
Start on the free tier or the cheapest plan and prove it works on one real task before you scale. Treat the first month as a test, not a contract.
How to commit safely
Run one real task end-to-end on the cheapest tier. If it saves time and the output is usable, expand. If not, the box was wrong for you — try another, not more spend. The businesses that win at AI are the ones with a rhythm of small, measured wins.
Done when: you've tested your box on one real task and decided whether to expand.
Two things to keep in mind
Your data matters. Know where your data goes before you commit. If you feed customer lists or financial records into a tool, check what it does with them. If you want AI on your own machines instead, that's the run-it-yourself box.
You don't have to do this alone. If you'd rather skip the trial and error, or you've hit a decision you want solved properly, that's the kind of thing I build for a fixed fee. The resources here are free for you to use either way.
See the four ways → pick your box → check the cost and reliability → commit small. One box, one small test, one measured win. That's the whole path.
