Backwards: AI Went Usage-Based While Legacy Software Went Flat-Fee
Two pricing migrations are happening at the same time, in opposite directions, and almost nobody is talking about how odd that is.
AI companies — the newest, fastest-growing, most capital-hungry software businesses in the world — priced themselves per token, per request, per unit of work. Meanwhile a long list of legacy vendors whose products have been metered for decades are marching toward flat annual platform fees.
The new thing charges you for what you use. The old thing wants a fixed cheque regardless. That is exactly backwards from how software pricing is supposed to mature.
- AI is metered because its marginal cost is genuinely large — compute is not free.
- Legacy vendors are fleeing metering because usage is now flat or falling.
- A flat fee on declining usage is a way to hide shrinking consumption inside stable revenue.
- For the buyer, the two models transfer risk in opposite directions.
- Watch which of your vendors wants to stop counting. It tells you what they see.
Why AI is metered
This one is easy and honest. Inference costs real money every single time. There is no version of an AI business where a heavy user and a light user cost the same to serve. Metering is not a pricing strategy, it is arithmetic.
It also has a property buyers should appreciate: it scales down. If we stop using it, we stop paying. Very little enterprise software has offered that in twenty years.
Why the incumbents are running the other way
Here it gets more interesting. If your product is metered and consumption is growing, metering is wonderful — revenue grows without a sales conversation. Vendors love that.
So when a vendor with a metered product suddenly wants to sell you a flat platform fee, the most likely explanation is that the meter has stopped being their friend.
| Model | Who carries the risk | What it signals |
|---|---|---|
| Usage-based | The customer | Vendor expects consumption to grow |
| Flat platform fee | The vendor | Vendor expects consumption to stall or fall |
| Per seat | The customer | Vendor is indexed to your headcount, which you are trying to cut |
| Minimum commit plus overage | The customer, twice | Vendor wants the floor of flat with the upside of metered |
A vendor moving from metered to flat is not simplifying your billing. It is buying certainty from you, and the price of that certainty is that you can no longer save money by using it less.
What this means at renewal
The flat fee is often pitched as a favour: predictable budgeting, no bill shock, simpler procurement. Sometimes it genuinely is. But it removes the one lever an efficient operator has — the ability to consume less and pay less.
If your team has just gotten dramatically more efficient, a flat platform fee locks in your old consumption level as a permanent floor. You did the work and the vendor kept the savings.
- If we cut usage in half next year, what happens to this invoice?
- Is the meter being removed because our consumption is falling?
- What is the true unit here — a seat, a request, or an outcome?
- What is the exit cost if we simply stop?
Where this settles
Our guess is that outcome pricing eventually wins in the middle — vendors charging for a result rather than a seat or a token. But that requires vendors to accept accountability for whether the thing worked, which most are not eager to do.
Until then, the tell is simple. Watch which vendors want to stop counting. They are telling you what they expect the count to do.
LAMPWORK operating analysis. Reflects our own vendor negotiations and observed market pricing at time of writing.
Own a brand this applies to?
We talk to founders at every stage — long before they're ready to sell.
Start a Conversation →