How large should a commitment be?

The level you would still reach in a disappointing year. Usage above it generally earns the same rate anyway, so a larger commitment buys nothing beyond the risk of a shortfall, and disappointing years are more common than the plan assumes.

What is the term to read first?

What happens to unused commitment, because that single term decides what a wrong forecast actually costs you. Rolling forward, expiring and simply being invoiced are three very different outcomes, and which one applies is settled at signature rather than discovered at the end of the period.

What is the risk nobody prices?

Being tied to one supplier while the field moves. Prices here have tended downwards and better options keep appearing, so a long commitment is partly a bet against both continuing, and that bet appears on no invoice or renewal notice.

Glossary

Committed spend

An agreed minimum you undertake to spend over a period, usually exchanged for a better rate than you would otherwise be offered.

In plain terms

Promising to spend a certain amount over a year in exchange for paying less per unit. The discount is real and immediate. The promise is a guess about your own future, made in a field where the sensible thing to do keeps changing.

Why it matters

Because the discount is easy to evaluate and the commitment is not, so the decision gets made on the half that is legible. What you are agreeing to is a constraint on future choices, and that constraint is worth pricing even though no invoice shows it.

How it works

The exchange is a better rate for a guaranteed floor. That is a reasonable trade and it is not free: the supplier is buying certainty from you, and the discount is what they are prepared to pay for it.

Size it against the volume you are confident about, not the volume you expect. Anything above the committed level still gets the improved rate in most arrangements, so the only thing a larger commitment buys is exposure if you are wrong.

The risk that costs more is not under-consuming. It is being tied to one supplier through a period when prices fall, a better model appears elsewhere, or the work you committed for turns out not to need this tool at all.

Prices in this area have generally moved downwards, which makes a long commitment a bet against that continuing. That is worth naming as the bet it is rather than treating the rate as fixed and the volume as the only variable.

What happens to unused commitment is the term to read first. Whether it rolls forward, expires, or is simply invoiced decides how much a bad forecast actually costs, and it is settled at signature rather than at the end of the period.

The workload matters as much as the number. Committing for something already running steadily is a forecast about a known quantity; committing to fund an expansion that has not happened yet is a forecast about a plan, and plans of that kind slip more often than they do not.

Two ways a commitment goes wrong

The asymmetry exists because one of these produces a number somebody has to explain and the other produces nothing at all. A shortfall shows up in a review, is attributable to whoever made the forecast, and is discussed. Paying more than the market rate for eight months, or continuing on a model that has been superseded because switching would strand a commitment, produces no artefact and no conversation. Nobody calculates what a different choice would have cost, so the loss is never counted, which makes it feel smaller than a shortfall a tenth its size. The practical protection is not to avoid commitments, which are frequently good value for steady work, but to keep the term short enough that being wrong is survivable and to size the floor against a bad year rather than a good one. Both of those cost a little discount and buy back the ability to change your mind, which in this field is the more valuable of the two.

Seen in the wild

A commitment sized against a planned rollout that then arrived a quarter late.
OpenRouter
A steady overnight workload, which is the case a commitment genuinely fits.
Make
A better and cheaper model appearing mid-term with the spend already promised.
Hugging Face

Common misconceptions

People assume

The main risk is not using it all.

In fact

That is the visible risk and rarely the expensive one. Being tied to a supplier while prices fall or a better option appears costs more than a shortfall, and it does not show up as a line anywhere.

People assume

A bigger commitment is better value.

In fact

Volume above the committed level usually attracts the same improved rate, so committing beyond what you are confident about buys exposure rather than savings. Size it to the floor you would hit in a disappointing year.

Questions

How large should a commitment be?: The level you would still reach in a disappointing year. Usage above it generally earns the same rate anyway, so a larger commitment buys nothing beyond the risk of a shortfall, and disappointing years are more common than the plan assumes.
What is the term to read first?: What happens to unused commitment, because that single term decides what a wrong forecast actually costs you. Rolling forward, expiring and simply being invoiced are three very different outcomes, and which one applies is settled at signature rather than discovered at the end of the period.
What is the risk nobody prices?: Being tied to one supplier while the field moves. Prices here have tended downwards and better options keep appearing, so a long commitment is partly a bet against both continuing, and that bet appears on no invoice or renewal notice.