Consumption pricing sells itself on flexibility. You pay for what you use, the bill scales with the value you get, and nothing is wasted. That story is true on the standard API, where every token you do not send is a token you do not pay for. It stops being true the moment you sign a committed spend agreement, because a commitment converts that flexible consumption into a floor you cannot fall below. This is the consumption trap, and it catches careful buyers as often as careless ones, because the trap is built into the structure of the deal rather than hidden in the fine print. Here is how it works and how to stay out of it.
On a pure consumption basis, your cost tracks your usage in both directions. Usage doubles, cost doubles. Usage halves, cost halves. That symmetry is the whole appeal. A committed spend agreement breaks the symmetry on the downside. You commit to spend a certain amount over a term, in exchange for a discount, and that commitment is a minimum. If your consumption rises above it, you pay for the extra, usually at a negotiated overage rate. If your consumption falls below it, you still owe the full commitment. The discount you accepted on the way in is the price of giving up your downside flexibility, and most buyers do not price that trade carefully.
The trap springs when consumption falls below the floor, and it falls for reasons that are usually good news everywhere except on this contract. You optimize the workload and it gets cheaper to run. A product line you forecast does not ramp as fast as planned. A team migrates some traffic to a cheaper model. Each of these is a win operationally and a loss against the commitment, because the savings you worked for disappear into a floor you already paid for. You did the right thing technically and the contract punished you for it.
The consumption trap is not the result of a predatory clause. It is the result of an honest forecast made at the wrong moment with the wrong incentives. When you are negotiating a commitment, the vendor wants a larger number because a larger commitment is more revenue and a stronger relationship. You want the discount that comes with committing more. Both pressures push the commitment up. Almost nothing in the room pushes it down, because the cost of overcommitting does not appear until later, when consumption comes in under the floor and the unused commitment quietly expires.
The forecasting itself compounds the problem. Most teams size a commitment off recent consumption extrapolated forward, often with optimistic growth baked in because the business case for the AI investment assumed growth. They rarely model the downside: the chance that optimization, slower adoption, or a workload change pulls consumption below the line. A commitment sized off the optimistic case and signed without a downside model is a commitment built to overshoot, and the consumption trap is what overshoot feels like on the invoice.
The sharpest edge of the trap is the treatment of unused commitment. On most Anthropic agreements, commitment is use it or lose it within the term. Spend you committed to but did not consume does not roll into the next period and does not come back as a credit. It simply expires. So a buyer who commits to more than they use is not deferring that spend, they are forfeiting it. The gap between the commitment and the actual consumption is pure loss, and on a large agreement that gap can run to seven figures over a term.
This is why the unused commitment terms deserve as much negotiating attention as the headline discount, and usually get far less. A modest improvement in how unused commitment is treated, a rollover allowance, a carryover window, a true down provision, can be worth more than another point or two off the rate, because it protects you in exactly the scenario the trap is designed around. Buyers fixate on the discount because it is the number on the front page. The protections that matter most are the ones that govern what happens when the forecast is wrong, and the forecast is almost always wrong in one direction or the other.
Avoiding the consumption trap is a matter of sequence and structure, and neither is complicated once you name them.
The most expensive version of the trap comes from committing against an unoptimized run rate, then optimizing afterward and watching the savings vanish into the floor. Reverse the order. Audit the workload, apply the routing, caching, and batch levers, let consumption settle at its real efficient level, and only then size the commitment against that clean number. The optimization is worth doing on its own, and doing it first means you commit to what you actually need rather than to padding.
Commit to a number you are confident you will exceed, not one you hope to reach. It is almost always better to commit conservatively and pay some overage than to commit ambitiously and forfeit unused commitment, because overage at a negotiated rate costs less than commitment you never use. Where the relationship is expected to grow, structure a ramp that starts low and steps up over the term, so the commitment tracks adoption rather than leading it.
Treat the overage rate, the unused commitment treatment, and any true down or carryover provision as first class terms, not afterthoughts. These are what govern the trap, and they are negotiable. A deal with a slightly smaller headline discount and strong downside protection often beats a deal with a bigger discount and a hard floor, once you account for the realistic chance that consumption comes in under plan.
Consumption pricing is flexible until a commitment turns it into a floor, and the consumption trap is what happens when your usage falls below that floor for reasons that should have saved you money. The trap is structural, not deceptive, which is why it catches careful buyers. Stay out of it by optimizing before you commit, sizing to the conservative case and ramping into more, and negotiating the downside protections as hard as the discount. Our pricing playbook lays out the commitment bands, the overage mechanics, and the protections worth fighting for. Download it and size your commitment against the real number.
Download the pricing playbook for the commitment bands, the overage mechanics, and the downside protections that matter.
Get the playbookWeekly intelligence on Anthropic pricing moves and the buyer side counters that work.