Author
TL;DR AI credits are a minted currency, and consumer games ran that experiment first. Three of its mechanisms transfer to B2B software: the unit decouples buying from spending, the burn rate rather than the unit price is where the effective price lives, and bundles turn prepayment into revenue with no delivery attached. The famous one, the whale, does not transfer at all. What arrives instead is an evaluation problem that surfaces at renewal.
AI credits borrowed their design from video games. Not the plumbing, the design: a vendor-defined unit standing between the customer’s money and the thing they actually buy. The moment you charge in a unit you define, you have minted a currency, and three powers come with it whether you wanted them or not. You set the conversion rate between your unit and money. You set the burn rate between your unit and work done. And you can change either one without ever changing a published price.
Gems, credits, tokens, compute units, points. The name varies; the decision does not. Consumer software ran this experiment for roughly two decades, from arcade tokens through mobile free-to-play, and reached conclusions B2B software is now rediscovering from scratch. Three of those conclusions transfer cleanly. The most famous one does not.
Minting a Currency Is a Pricing Decision, Not an Implementation Detail
Credit systems are usually adopted as billing conveniences: prepaid balances are easy to support, and a credit smooths task-level cost variance into one number. All true, and all beside the point. A vendor that mints a unit owns a currency, and currencies behave like currencies whatever the issuer intended.
The unit itself is defined elsewhere: a credit is a surrogate unit, a vendor-defined billing unit that folds several underlying value metrics into one accounting layer, and what a credit in AI pricing actually is covers the mechanics in full. This piece is about the consumer branch of the family tree, where the commercial consequences were worked out at scale and in public. Most AI software pricing debates treat credits as a new problem. The problem is older than the category.
What a Minted Unit Does, in Any Industry
A minted unit does three things, each a property of the unit rather than of the industry around it.
The first: the unit decouples the purchase from the price. Buying the currency and spending it become two acts, separated in time. Casino chips are the cleanest illustration: there, the decoupling is the entire design intent, and nobody pretends otherwise. Once the unit is in hand, the spending decision is no longer a money decision. The player left their wallet at the cage. Your customer left theirs at the annual order form.
Why the burn rate is the real price list
The second mechanism: the burn rate is the real price list, and it is the part nobody publishes. A rate card denominated in credits tells a buyer nothing about cost until they know how many credits a real piece of work consumes. Different actions burn at different rates, the vendor sets the mapping, and the mapping is where the effective price lives. This is how a credit price list stays perfectly stable while the price moves underneath it. It is also why no two vendors’ units can be compared: Atlassian sells AI work in credits while HubSpot sells it in resolutions, and there is no exchange rate between them.
A bundle discount and a prepayment inducement look identical on a pricing page
The third: bundles teach volume, and unspent balance is revenue with no delivery attached. Larger packs carry a lower per-unit price, which reads as a volume discount and functions as a prepayment inducement. The expiry windows and forfeiture terms around the unspent balance are the subject of credit expiration and breakage, and the structural consequences of running this design in B2B are catalogued in the six flaws of credit-based pricing.
None of this is a scandal. These are consequences of the design, and a vendor that names them can engineer around every one. The vendor in trouble is the one that has not noticed them, because that vendor is being priced by its own accounting layer.
The One Thing That Does Not Transfer
The famous fact first, stated fairly: in free-to-play games, a small share of players carries most of the revenue. The industry’s own analytics reporting documents the concentration, and the figures vary widely between reports. It is real. It built an industry.
Why there is no B2B whale
It does not transfer, for structural reasons. The whale exists because the spender is an individual with unbounded discretionary budget, no approval step between impulse and purchase, and nobody reading the statement. Remove any one of those conditions and the concentration weakens. B2B software removes all three. There is a budget owner. There is an approval chain. There is a finance team whose job is reading the statement. B2B revenue concentrates too, but for an unexciting reason: enterprise accounts are larger. That is a fact about account size, not behavior.
So the conclusion is the uncomfortable one. The free-to-play design imported into B2B loses the mechanism that made it lucrative and keeps the mechanism that makes it hard to evaluate. A vendor adopting credits is not acquiring a whale strategy. It is acquiring an opacity problem.
The decoupling lands on procurement, then on renewal
In a game, the decoupled unit lands on an individual who at least knows they are spending. In B2B it lands twice: on procurement, asked to approve a spend it cannot convert into a unit of work, then on finance at renewal, reconstructing the effective price from a year of invoices and frequently not liking the answer. Both are durability problems, and durability is where pricing architecture is judged.
In a game, the decoupled unit separates a player from their money. In B2B, it separates a buyer from their ability to evaluate the deal, and that bill arrives at renewal.
Where Does Your Pricing Architecture Actually Stand?
A few questions return your pricing architecture score and show which of your licensing, packaging, and pricing decisions needs attention first. Real diagnosis, not a mailing-list toll.
Regulators Have Already Read This Design Once
In March 2025, the EU’s Consumer Protection Cooperation Network adopted key principles on in-game virtual currencies. The headline requirement matters here: show the price of digital content in real-world currency even when the transaction runs through a virtual currency, and do not use the currency layer to obscure what something costs. The principles are soft law, not binding regulation. Their significance is directional. Consumer protection authorities looked at a minted unit, in the setting where it was invented, and concluded that the decoupling itself was the consumer harm worth naming. Not the prices. The layer.
Do consumer protection principles govern B2B software contracts? No, and nothing here implies they do. What transfers is the reasoning. An enterprise procurement team applies the same test on its own authority: can we see what this costs in money, per unit of work, before we sign? And it asks with more leverage than any consumer has ever held.
Which is why the vendor’s real exposure is commercial, not regulatory. The buyer who cannot price the unit at purchase will price it at renewal, with a year of invoices in hand and considerably less goodwill.
Tests for Your Own Minted Unit
The lineage is only useful if it changes the design review. Five questions, answerable from your own published materials and ledger.
Can a buyer, using only what you publish, convert your unit into the cost of a piece of work they actually do? If the answer requires a call with your sales team, the unit is doing pricing work you have not designed.
When your LLM inference costs move, which lever do you reach for: the price of the unit, or the number of units a task consumes? If it is the second, you have a price change no customer will see announced and every customer will eventually feel, rationed back as a token diet.
If you published your burn rates tomorrow, what would break? A design that only works unpublished is a design with a renewal problem on a timer.
Does anything in your business distinguish revenue from units sold and revenue from units consumed? If those two numbers have never been separated, some part of your growth is a balance sheet artifact.
Would your largest account be able to reconstruct its effective price per unit of work from its own invoices? They will attempt it. The only question is whether they do it with you or without you; run that reconstruction with a pricing expert first, on your side of the table.
The argument is not that minting a unit is wrong. A minted currency is a design whose consequences another industry has already paid to discover. The choice is whether to inherit the conclusions or repeat the experiment.
Credits, gems, and compute units answer to the same design, and the design has a longer track record than the AI category now using it. If your unit would not survive its own publication, talk to an expert: describe what the unit maps to and where the burn rates live, and a pricing expert replies with where the fault line sits.
