Oracle is looking to grow revenue from existing customers as AI embedded in its enterprise software handles more work. At its Investor Day on October 28, 2026, Chris Leone, executive vice president of application development, explained that as the volume of AI work grows, consumption of AI units will rise and revenue can grow without increasing the number of contracted users, The Register reported. The strategy adds usage-linked AI revenue on top of traditional per-user subscription income.
But the explanation that charges arise in line with AI workload does not reveal whether the model bills for usage after the fact or draws down a quota purchased in advance. Cross-checking the price list Oracle published on October 8, 2026, against its contract document "Fusion Cloud Service Descriptions v102026" (effective October 20) shows that under three plans (direct AI unit purchase, per named user, and per employee), a fixed number of AI units is granted up front. Dividing the list price by the units granted gives $0.01 per unit under all three. What determines actual usable capacity is the number of units granted under the contract and the consumption multipliers set for each AI model and type of work.
Oracle's AI Revenue Strategy and the Four Contract Types on the Price List
At the October 28 Investor Day, Oracle presented a long-term financial plan premised on growth in its AI-related business. According to an exhibit to a Form 8-K filed with the U.S. Securities and Exchange Commission (SEC), the company expects annual revenue of at least $225 billion in fiscal 2030 (the year ending May 2030). It also said it would limit additional funding for its investment plan from fiscal 2027 through fiscal 2030 to $20 billion, excluding refinancing of existing debt. As part of this growth plan, the company is trying to turn AI use in enterprise applications into a new revenue source.
On the product side, Oracle announced "Oracle Fusion Claw," an execution foundation for agentic applications, on September 29. It combines AI reasoning and planning with conventional computing to carry out complex tasks autonomously. According to Oracle's announcement, 25 applications that use Fusion Claw were added, bringing the company's total of agentic applications to 75. They cover tasks such as accounting, HR, logistics, and sales.
Charges tied to these AI features are listed in the "Fusion Cloud Service Global Price List" that Oracle published on October 8. The price list carries the following four SKUs (units of sale).
| Contract / service | List price (USD) | Billing unit | Minimum order |
|---|---|---|---|
| Fusion AI Units Cloud Service | $1,000 | 100,000 AI units | 1 unit |
| Fusion AI Agents Cloud Service (named user) | $150 per month | 1 named user | 50 users |
| Fusion AI Agents Cloud Service (employee) | $5 per month | 1 employee | 500 employees |
| Fusion Agentic Applications Cloud Service | $500,000 per year | 1 application contract | 1 contract |
All of these are public list prices before discounts and do not show actual contract amounts. What matters is that AI unit allowances are included not only in the direct-purchase model but also in contracts based on the number of users or employees. That said, the four contract types do not all share the same pricing structure. The annual agentic applications contract includes rights to use the applications in addition to the AI unit grant.
$150 a Month per Named User Buys 15,000 Units, and the Effective Price Is the Same Across Three Plans
In Oracle's contract document "Fusion Cloud Service Descriptions v102026", the number of AI units granted is defined for each plan. In a named-user contract, 15,000 units × number of named users × months of the contract term are granted in a lump sum at the start of the term; in an employee-based contract, it is 500 units × number of employees × months. When AI units are purchased directly, 100,000 units are added to the allowance per purchased unit. In every case, the units set by the contract are secured first, and the balance falls as AI features are used.
Calculated from these grants, the per-unit price is identical across the three plans. Buying 100,000 units for $1,000 gives 1,000 ÷ 100,000; $150 per named user per month gives 150 ÷ 15,000; and $5 per employee per month gives 5 ÷ 500. All come to $0.01 per unit. However, this is a conversion figure obtained by dividing the public list price by the units granted, and it does not mean that the usage rights or user restrictions attached to each contract are identical.
Using the European Central Bank (ECB) reference rate of $1 = ¥158.06 on October 29, 2026, one unit is about ¥1.58, $150 per named user per month is ¥23,709, and $5 per employee per month is about ¥790. Because the named-user contract has a minimum of 50 users, the minimum contract value at list price is $7,500 per month. These are reference conversions using an exchange rate and do not show what Japanese customers would actually pay.
The $500,000-a-year "Fusion Agentic Applications Cloud Service," meanwhile, has to be considered separately from the other three. This contract grants an allowance equivalent to 2.5 million units per month, or 30 million units a year for a 12-month term. At $0.01 per unit, that is worth $300,000, or 60% of the annual contract price. That leaves a $200,000 gap, so the full $500,000 cannot be treated as payment for AI units. However, Oracle has not explicitly allocated this difference to individual feature fees or the like, and the breakdown of the price cannot be determined.
The contract terms also confirm that the three plans grant the allowance first. Oracle's contract document provides that purchased AI units not used by the end of the term expire, with no refund or fee reduction for unused portions. If the balance runs short, the customer must buy additional units or stop using features that consume units. If the balance falls below zero, Oracle reserves the right to restrict, reduce, or block access to the relevant features.
Importantly, the AI-unit provisions reviewed contain no clause that automatically bills after the fact for usage exceeding the allowance. "Prepaid allowance" here does not refer to the actual payment date; it means a mechanism in which the number of units usable during the term is granted first, and the customer must buy more or stop use once they run out. Named-user and employee contracts are listed as monthly fees on the price list, but the document alone does not show the specific timing of payment. AI use in development and test environments also counts toward unit consumption, so work other than production operation must be included in managing the allowance.
Consumption Multipliers That Turn the Same AI Task from 0 Units into 20
AI unit consumption is not determined simply by the number of executions. Oracle's contract document uses a method in which the input and output tokens of each task are added together, divided by 10,000 and rounded up to a whole number, and the result is multiplied by a consumption multiplier (Action Value Factor) set by model tier and type of work. A token is the unit AI uses when processing text and the like, so even the same task consumes different amounts depending on the input and the length of the response generated.
For example, if combined input and output is within 10,000 tokens, the value before applying the multiplier is 1. If the total reaches 10,001 tokens, it is rounded up to 2, so a model with a 20x multiplier consumes 40 units. Unit consumption is not strictly proportional to token count; note that it is rounded up in 10,000-token increments for each task.
Table 1 of the contract document sets the following multipliers by task and model tier.
| Type of work | Basic tier | Frontier tier |
|---|---|---|
| General tasks (answering questions, summarization, transaction approval, reasoning steps) | 0x | 20x |
| Fusion Claw tasks (in-depth research, autonomous execution, planning) | Not available | 20x |
Based on Table 1 of Oracle Fusion Cloud Service Descriptions v102026. Consumption multipliers took effect on October 1, 2026. The contract document also includes a Balanced tier and a Bring Your Own (customer-supplied model) tier.
Under this scheme, even a general task within 10,000 tokens consumes 0 units on a Basic-tier model but 20 units on the Frontier tier. At list price, 20 units is $0.20, or about ¥32 at the exchange rate above. If a single task totals 1 million input and output tokens, the Frontier tier consumes 100 × 20 = 2,000 units, which is $20 (about ¥3,161) at list price. However, splitting work into multiple tasks causes rounding up for each task, so consumption can increase even for the same total number of tokens.
Which AI models fall into which tier is set out in Table 2 of the contract document. The Frontier tier includes Google's Gemini 3.5 Flash and OpenAI's GPT-5.5, while OpenAI's GPT-OSS running on Oracle Cloud Infrastructure (OCI) is classified as Basic. Cohere Command-R-08-2024 and Meta Llama 3.3 are in the "Included" tier and do not consume AI units for eligible use.
Oracle describes GPT-OSS in the Basic tier as suited to development, testing, and basic uses with relatively low volume. For AI agent processing involving planning and tool calls, it recommends models other than Basic. In fact, Fusion Claw tasks cannot use the Basic tier; a 20x multiplier applies to the Frontier tier and 10x to customer-supplied models. The models Oracle cites as the reasoning foundation for Fusion Claw are also frontier models from Google and OpenAI.
For this reason, even if the volume of AI work grows, not all processing consumes AI units at the same pace. Processing that fits within the free allowance or uses a model with a 0x multiplier does not lead to additional unit purchases. On the other hand, if complex tasks using the Frontier tier increase, the allowance falls faster. The revenue growth from expanded AI use that Oracle described at Investor Day depends on this mix of tasks and model choices.
In addition, under the contract Oracle can change the table of consumption multipliers and eligible models for reasons such as changes in third-party model pricing, changes in underlying technology, or the addition of new services. Changes generally require at least 30 days' notice and apply to future use after the notice period. In other words, even if the allowance within a contract term stays the same, the number of units needed for the same task could rise or fall if model classifications or multipliers change. When estimating AI operating costs, companies need to consider the risk of such multiplier changes, not just the unit price.
A Standard Allowance of 30,000 Units a Month and the Claim of "No Extra Charge for AI"
In thinking about Oracle's AI pricing, the relationship to the company's earlier stance of "not charging extra for AI" also matters. According to Computing's reporting, Steve Miranda, executive vice president for applications, said at Oracle AI World Tour in London that the company would not impose an extra charge for AI. Forrester's analysis published in May also reported that Oracle repeated at an April analyst briefing that it did not plan to charge extra for AI agents embedded in Fusion. These statements may seem at odds with the AI unit pricing that has now come to light.
However, Oracle's contract document defines a standard allowance granted separately from purchased AI units. Specifically, 30,000 units a month are allocated per Fusion production environment, and consumption from AI use is deducted from this allowance first. Only after the standard allowance is used up is consumption deducted from the shared allowance purchased or granted under the contract.
At the public list price of $0.01 per unit, 30,000 units is equivalent to $300 (about ¥47,418). For a general Frontier-tier task with input and output of 10,000 tokens or less per run, each run consumes 20 units, so in theory that is 1,500 runs a month. However, this standard allowance is also consumed by other AI processing tied to the same production environment, so no particular agent is guaranteed 1,500 runs. And general tasks using the Included tier or the Basic tier with a 0x multiplier do not reduce AI units at all.
Given this structure, Oracle's claim that it will not charge extra for AI and a system requiring purchases of AI units according to usage are not necessarily contradictory. It can be read as meaning that AI features built into existing Fusion applications can be used at no additional charge, while use beyond the standard allowance, and advanced agent features that require a separate contract, incur costs. Which features are included in the standard contract and which require an additional one has to be checked against individual services and contract terms.
At the same time, even under a method of buying the allowance first, the need to buy more each time it runs out supports the view that it is effectively usage-based pricing. Indeed, Leone's explanation as relayed by The Register was that Oracle's revenue grows as the volume of work AI handles increases. This analysis therefore does not claim that AI usage has no effect on customers' total payments. Its point is that the billing mechanism that can be confirmed from the public contract documents consists of a pre-granted allowance and additional purchases, rather than automatic after-the-fact billing for excess use.
For Fusion Claw in particular, the unavailability of the Basic tier matters on the pricing side. Oracle itself shows a configuration using Frontier-tier models for complex agent processing, which consumes 20 units even for tasks within 10,000 tokens. The more an AI agent repeats planning and reasoning, the more the standard 30,000 units and any purchased allowance are drawn down. For Oracle, this means AI usage growth can translate into revenue without adding users; for customers, it means understanding per-task unit consumption and judging when additional purchases will be needed.
Miranda has also mentioned the possibility that software pricing could in future shift from a user-count basis to one based on transaction volume or company size. The AI units do not immediately abolish the existing per-user pricing, but they are positioned as a mechanism that adds the volume of work AI performs as a new revenue metric. It can also be seen as a shift toward measuring the value of enterprise software not only by the number of employees who use it but by the amount of work AI actually processes.
What will shape the interpretation of this pricing going forward is how Oracle revises its contract document and price list. If the current "Fusion Cloud Service Descriptions v102026" or "Global Price List" is amended to add a clause that automatically bills after the fact for AI unit shortfalls, or if the named-user and employee SKUs are changed to billing based on actual consumption, this reading of a system that grants an allowance first would break down. Conversely, if no such changes are made, Oracle's goal of expanding revenue in line with AI usage is likely to proceed for the time being through upfront grants of allowances and additional purchases.