Token Discounts: The Enterprise AI Contract Shifts Its Center of Gravity

The facts: the discount dies with the last token

On Monday 28 September 2026, The Information published a contractual detail that carries more weight than any benchmark. Anthropic ends the discount for enterprise customers the instant the token volume set out in the contract is used up.

The figure at the heart of the story comes from managers at software companies that buy from both providers: roughly 15% off list price, according to the account by Kevin McLaughlin, picked up by The Next Web^[1]. Past the threshold, the customer either renegotiates or moves to full price. The original report sits behind The Information^[2], and TNW notes that it reported the story without independent verification.

The token is the unit these providers use to measure the text their models read and write. Translated into procurement terms: your unit of consumption changes price halfway through the fiscal year.

What this clause really is

Commercial language presents it as an ordinary commitment threshold. In substance it is a renegotiation lever, placed at the exact point where the customer has already moved workloads into production.

Anyone who exceeds the contracted volume has already embedded the model in their products, trained their teams, rewritten their workflows. At that moment the bargaining power sits with the provider, and the price says so.

OpenAI applies a softer variant: the remainder of the current month plus one additional month to close the new agreement, according to a licensing consultant quoted in the report. Then list price kicks in. The difference looks cosmetic, and instead it measures exactly how much room the CFO has to manoeuvre when consumption grows beyond forecast.

From benchmark to clause: the competitive shift

For two years the race between model providers has been fought over scores, context windows and latency. This story moves the axis: competition now runs through the text of the contract.

The moat for enterprise AI providers lives in deployment and commercial terms, more than in the model. A customer who has built a product on top of an API will accept an upward renegotiation to protect the roadmap. The provider with the tighter clause captures more durable revenue than the provider with the higher score.

Frederick Philipson, co-founder of Redress Compliance, describes OpenAI as the most aggressive player on discounts right now, in the middle of a hard push on enterprise sales. In the same month, OpenAI hired the first global head of sales in its history. Two signals pointing in the same direction: the battle is being fought inside the buying cycle.

The cloud providers are playing a different game

Amazon, Microsoft and Google treat overage the opposite way. They typically leave the customer the discount even on consumption beyond the commitment, for the full term of the contract, explains Jeff Muscarella, chief innovation officer at NPIFinancial, in the same report.

That is the distance between selling capacity and selling a multi-year relationship. The hyperscalers have twenty years of enterprise habits behind them, and those habits now become a sales argument against going direct to the labs.

For the Chief Digital Officer the consequence is concrete. Buying a model through the cloud channel or through a direct contract changes the risk profile of overage, with the technology held constant. Choosing the channel becomes a financial decision before it is an architectural one.

Share-of-wallet clauses and the cost of lock-in

A second element appears in the report, relayed by two managers: both providers have begun inserting share-of-wallet clauses into discount agreements. The large customer commits to directing the bulk of its AI budget toward a single platform.

This is the real news for the board. The discount becomes the price of exclusivity, and multi-vendor strategy turns into an explicit line item.

Keep two providers active and you pay more at list price while gaining bargaining power at renewal. Concentrate and you get the better price today while handing the provider control of the next negotiation. That choice belongs to the Chief Strategy Officer, on a three-year horizon.

The evidence that customers do move

Harjot Gill, chief executive of CodeRabbit, has said that OpenAI is now his company's primary AI provider. Six months ago that place belonged to Anthropic.

The company is three years old and spends tens of millions of dollars a year on AI, again according to his own words in the report. A single case, with a precise value: at these spending levels, the cost of switching remains payable.

The scale of the business helps clarify what is at stake. In the twelve months ending in June, more than 100 companies spent over $10 million a year with Anthropic, and more than 1,000 passed the $1 million mark. Meta and Cursor are among the largest customers.

Anthropic's revenue overtook OpenAI's for the first time in the most recent quarter. A position of commercial strength explains a hard line on overage rather well.

List prices fall, contract terms tighten

In recent weeks both providers have cut list prices: Anthropic with a cheaper Opus model, OpenAI by halving GPT-6 prices. The frontier tier costs less with every release.

The model becomes a commodity, and value migrates elsewhere. Buying generic AI capacity means buying a deflationary good, while the line item that grows is compute and the constraint that bites is the contract structure.

The consequence for the CFO is clear. Savings on unit price evaporate when the overage clause returns the bill to full list price mid-year. The number to watch at budget time is the effective cost per token at year end, as distinct from the price written on the provider's slide.

What to decide in the next 90 days

Four concrete moves, before the next renewal.

• Map real token consumption by product and estimate the date the contracted volume runs out.
• Ask in writing how overage is treated: discount maintained, grace window, or immediate list price.
• Assess the cloud channel as a hedge against overage risk, with the model held constant.
• Quantify the price of exclusivity before signing a share-of-wallet clause.

The market has moved. Token negotiation has become the arena where AI providers defend their margin, and the board that arrives at renewal unprepared pays the difference in cash.

There is only one question to take to the next investment committee: what does your overage cost the day after you cross the threshold? Those with the answer in writing negotiate as equals. Those who find out on the invoice negotiate from weakness.