How much should AI tools actually cost?
I run a small AI product. That means I get an invoice every month for the AI usage my users generate, and I've spent an unreasonable amount of time working out what things actually cost. This post is what I've learned, written for people trying to figure out whether the $20/month they're paying is fair.
Short version: some AI pricing is genuinely reasonable, some is pure brand margin, and the difference is easy to spot once you know what to look for.
Why everything costs $20/month
Look at the AI landscape and you'll notice something odd: an enormous number of products are priced at almost exactly $20/month. ChatGPT Plus, Claude Pro, Gemini's paid tier, dozens of smaller tools.
That's not a coincidence, and it's not a cartel. It's anchoring. Once the two biggest players in a category settle on a number, everyone else prices near it, because being much more expensive requires justification and being much cheaper signals "worse".
The $20 price point tells you almost nothing about what the product costs to run. It tells you what the market got anchored to.
Which means when you're evaluating a $20/month AI tool, "is this worth $20" is the wrong question. The right one is "how much of this $20 is compute, and how much is brand".
What actually goes into the price
For any AI tool, the cost stack looks roughly like this:
| Cost | Roughly | Notes |
|---|---|---|
| AI provider fees | The bulk of variable cost | Scales directly with how much you use it |
| Servers / infrastructure | Small at low scale | A modest server handles a lot of users |
| Payment processing | ~2-5% | Higher for international / small transactions |
| Support | Varies enormously | Biggest hidden cost for consumer products |
| Marketing / acquisition | Often the largest line | Funded products spend heavily here |
| Margin | Whatever's left | Where investor-backed and bootstrapped differ most |
The line that surprises people is marketing. For a venture-funded AI product, customer acquisition cost frequently exceeds every other line combined. When you pay $20/month, a meaningful chunk of that is paying for the ads that found you.
This is why bootstrapped tools can often be cheaper for the same underlying capability. Not because they're more efficient at engineering — because they aren't spending $80 to acquire you.
The maths that actually matters: cost per useful output
Monthly price is a vanity metric. What matters is what one useful thing costs you.
Work out your real number:
- Count how many times you used AI last week. Be honest — most people guess 3x too high.
- Multiply by 4 for a monthly figure.
- Divide your monthly subscription by that number.
Some real examples of what that produces:
| Scenario | Monthly | Uses/mo | Cost per use |
|---|---|---|---|
| Heavy daily user on $20 plan | $20 | 600 | ~$0.03 |
| Moderate user on $20 plan | $20 | 80 | $0.25 |
| Occasional user on $20 plan | $20 | 15 | $1.33 |
| Moderate user on $6 plan | $6 | 80 | ~$0.075 |
The occasional user paying $1.33 per interaction is the one being quietly fleeced — not by a scam, but by a plan designed for someone else's usage pattern. If that's you, you don't need a better tool. You need a smaller plan or a short pack.
Why almost nobody sells short packs
Here's something I learned building this: the pricing model that would suit most casual users barely exists. Almost every AI tool sells monthly subscriptions and nothing else.
The reason is boring. Recurring revenue is what investors value, what makes a business predictable, and what survives a slow month. A $1 three-day pack produces none of that. It produces a customer who might come back in eight weeks.
I sell one anyway, because I think a lot of people genuinely want to test a tool properly for a few days without committing to a subscription they'll forget to cancel. It's not a growth-optimal decision. It's just what I'd want as a user.
Five signs you're overpaying
- You have three overlapping subscriptions. Most people I talk to about AI costs are paying for two tools that do the same job. Cancel one this week.
- You can't remember the last time you hit a limit. If you never approach the ceiling, you're on too big a plan.
- You're paying for features you've never opened. Image generation, custom bots, team seats. Check what you actually touch.
- You're on annual billing for something you've used twice. Annual discounts are only a discount if you use the thing.
- You haven't checked regional pricing. Many vendors price significantly lower in India, Brazil, Indonesia and elsewhere. Worth thirty seconds of checking.
What I actually think is fair
My honest view, as someone who sees both sides of the invoice:
- $0 — Free tiers should be genuinely usable, not demos. If a "free" tier gives you five messages, it's a trial and should be labelled one.
- $5-10/month — Fair for a focused tool with a real allowance. This is where most individual users should be.
- $20/month — Fair if you use it daily and it replaces real work time. Poor value if you use it twice a week.
- $40+/month — Only makes sense for teams, agencies, or people whose income directly depends on output volume.
The uncomfortable part: most consumers are on the $20 tier when the $6 tier would cover them. Vendors know this. The plan page is designed so the middle option looks like the sensible one.
The thing nobody tells you about AI costs
Underlying model costs have fallen dramatically over the past few years — significantly cheaper per unit of work than they were. Consumer prices have not moved.
That gap is margin, and it's mostly funding growth. It'll compress eventually, the way every category's does, but it won't happen because vendors get generous. It'll happen when enough users work out what they're actually buying and start choosing accordingly.
Which is a long way of saying: do the cost-per-use maths on your own subscriptions this month. It takes five minutes and most people find at least one thing to cancel.
— sane. If you've found a genuinely well-priced AI tool I've missed, tell me: hello@writerpi.com. More posts.