Author
TL;DR A dollar-anchored credit is still a surrogate unit. Pegging one credit to one cent or one dollar fixes the exchange rate between money and credits, and nothing else. The price sits in the conversion table, the vendor-authored rate sheet that says how many credits each action consumes, revisable at will and expandable with every new feature. Credit-system praise celebrates the peg and skips the table. Procurement reads the table first.
Dollar-anchored credit systems publish the same proud arithmetic: one credit equals one cent, and a $10 budget holds 1,000 credits. The transparency is genuine as far as it reaches, which is the exchange rate. AI credit conversion rates, the table stating what each action costs in credits, are where the price of the work sits, and the vendor writes that table alone.
The peg prices the credit. The table prices the work. Only one of those numbers appears in the praise.
The Dollar Peg Everyone Publishes
The pattern is spreading for a reason: it photographs well. GitHub’s Copilot move is the cleanest recent specimen we have tracked: plan prices stayed where they were while the unit beneath them changed from premium request units to a dollar-anchored credit. Our Pricing Observatory holds the dated record, captured from GitHub’s own announcements:
A currency-denominated credit makes the wallet legible. Set the standard Business allowance in that table beside the same seat’s price and the two are one number in different clothes, penny for penny. Finance can budget in dollars, admins can cap spend in dollars, and nobody needs a decoder ring to know what the balance is worth.
So credit the peg with what it fixes. In an unpegged system, the vendor holds two levers: the rate between money and credits, and the rate between credits and work. AI credits inherited both levers from the minted currencies of video games, and the dollar peg genuinely surrenders the first one. A credit that always costs one cent cannot be devalued at the checkout counter.
That still leaves the second lever. The second lever is the price.
AI Credit Conversion Rates Are the Price List
Every credit system carries two numbers. The first is what a credit costs in money, pegged and published. The second is what an action costs in credits: the conversion table the vendor authors, revises, and extends every time a feature ships. A buyer who knows the first number knows their budget. Only the second tells them what anything in the product costs.
The same GitHub change, the one that redrew Copilot’s position on the AI pricing spectrum, demonstrates both halves of the structure in one dated move. The day the one-cent peg took effect, model multipliers rose, and code review began drawing on two meters at once, GitHub Actions minutes alongside AI Credits. The currency held its value. The work repriced.
SAP’s AI Units run the same architecture at enterprise scale, and our Observatory record traces it back before the current AI wave made credits fashionable. In October 2023, SAP established AI Units as a prepaid cross-portfolio currency for premium Business AI: purchased annually, expiring after twelve months, pooled across products, while the contract’s formal usage metric remained the Capacity Unit. The published price list carried the currency. A companion services list carried the per-service conversion factors, where Joule was metered in Messages sold in 10,000-Message blocks at a factor of 7.
Read that structure the way a procurement team eventually will. The unit on the contract is stable and dignified. The number deciding what a month of actual usage costs sits in a companion document, per service, revisable by version. That is a surrogate unit doing exactly what surrogate units do: one currency in the buyer’s hands, many rates in the vendor’s.
What the Dollar Peg Buys the Vendor
None of this makes dollar anchoring a trick. It solves real problems. Being precise about which ones explains why vendors keep reaching for it.
The peg gives a multi-product portfolio one wallet, which SAP’s pooled AI Units and GitHub’s org-level budgets both deliver. And the conversion table absorbs the volatility of LLM inference costs: when a new model changes the economics underneath a feature, the vendor adjusts a multiplier rather than reprinting every price. That last property is the quiet reason the table stays in the vendor’s hands.
What the Vendor Conceded and What It Kept
Notice what the vendor conceded and what it kept. It conceded the exchange rate, which was never the margin lever. It kept the rate sheet, which is.
Why a Pegged Credit Is Still Not a Value Metric
A value metric is a unit the buyer can count without any vendor table: a resolved conversation, a seat, a completed run. A pegged credit still needs the table to mean anything, so the peg moves the unit no closer to value. It moves the opacity one layer down, from the currency to the rates.
SAP’s May 2025 restructuring shows the table carrying the whole pricing function while the currency sits still. Joule’s base capabilities became free, and monetization moved to premium per-user packages denominated in AI Units, banded from 8 down to 1 AI Units per user per month. System-triggered capabilities went to pure consumption pricing, down to 0.005 AI Units per record for document grounding. Every one of those numbers is a conversion rate. None of them is the price of an AI Unit.
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.
Renewal Arrives on the Table, Not the Peg
In our client work, the pattern repeats: a credit layer feels like margin protection at design time, then hands procurement its best lever at renewal. Once a buyer realizes the credit’s meaning is set by the conversion table, the table becomes the negotiation. Sophisticated procurement teams now skip the headline rate entirely and press where the price can move.
A vendor who treats a conversion-table change as a product update is discovering that experienced buyers treat it as the price change it is. The dollar peg sharpens this rather than softening it, because the arithmetic is now trivial for the other side too. When a multiplier doubles, the buyer can state the increase in dollars without help. The peg does the math for them.
What Separates a Defensible Credit System
The credit structure itself is not the defect. We have recommended engineered credit architectures where the criteria hold: a published conversion table, changes to what a credit buys treated as price changes with notice, and exportable underlying events. What separates a defensible credit system from the six-flaw version is that engineering, not the presence of a dollar sign in the peg. A pegged credit with a floating, unnoticed table is transparency theater: the visible half is fixed precisely because the operative half is not.
Why Credits Are Not a Starting Position
That is also the answer to the advice, now common from the venture side and from the platform vendors themselves, that an application-layer company should start with credits. Starting with credits inverts the order the architecture requires. A credit is a surrogate unit, and choosing one is a licensing decision about what grants and limits access. Make that choice before you know which unit your buyers can budget against and you have picked the accounting before the metric. Every property above that makes a credit system defensible depends on knowing what the credit stands for, and a company that has not yet learned what its customers are buying does not know.
Read the incentives on that advice before you take it. The loudest voices for starting with credits are usually the ones positioned downstream of the decision: billing platforms that meter credits, entitlement vendors that enforce them, and investors holding positions in both. None of that makes the advice wrong, and some of these are good products. It does mean the recommendation reaches you from people who are paid the same whether or not the structure survives your renewal. Ask anyone recommending a starting metric what happens to them if it turns out to be the wrong one.
Four Tests Before You Anchor Credits to the Dollar
If you are designing or defending a dollar-anchored credit system, put these four questions to it before your customers do.
- Can a customer learn what an action costs in credits from anything you publish, or only what a credit costs in dollars?
- If a conversion factor changed tomorrow, would any published price change with it, anywhere?
- Does your contract treat a re-rate of the table as a price change with notice, or as a product update?
- Who owns the conversion table, and does a pricing decision-maker review its changes before they ship?
A system that passes all four is defensible at renewal. A system that fails them has two price lists, one for the brochure and one for the invoice, and the gap between them is what a renewal negotiation will find. Answer them before a redesign rather than after an escalation, and if the answers are uncomfortable, take them to a pricing expert while the table is still yours to change.
The conversion table still decides what everything costs, and the vendor still holds the pen. If your credit architecture leans on that asymmetry, talk to an expert. Describe how your credits map to actions today and what you intend to charge for next year, and a pricing expert replies with where the structure holds and where it will be contested.
