# Oracle Cost Advisor for AI: Supported Providers, Availability and Setup

Oracle's new cost tool connects AI spending to finance workflows, with a Fusion EPM dependency that matters as much as the dashboard.

By Paras Madan | 2026-10-28


Oracle introduced Cost Advisor for AI in limited availability on October 27, initially in OCI's US East Ashburn region. It brings spending from OCI Generative AI, OpenAI and Anthropic into a shared view, then connects that information to budgeting and model-cost decisions through Oracle Fusion EPM.

## What Oracle Cost Advisor for AI does

The [release announcement](https://docs.oracle.com/en-us/iaas/releasenotes/generative-ai/cost-advisor-limited-availability.htm) describes provider comparisons, cost allocation, budgets, forecasts and recommendations for cheaper models that could offer similar quality. Estimated savings are opportunities to evaluate, not a promise that switching models will preserve results.

The relevant audience is an organization where several teams use different AI providers and the finance team needs a common account of that spending. An API bill can show what was consumed without explaining which department owns it or which business process produced the cost.

Cost Advisor tries to connect those views. It does not automatically know every provider an organization uses, and a dashboard can only describe the connections and mappings supplied to it.

## Supported providers and the Fusion EPM workflow

Oracle's [overview documentation](https://docs.oracle.com/en-us/iaas/Content/generative-ai/cost-advisor.htm) lists the supported sources more precisely.

| Provider | Supported services |
| --- | --- |
| Oracle | OCI Generative AI |
| OpenAI | OpenAI Platform and ChatGPT Enterprise |
| Anthropic | Claude Platform and Claude Enterprise |

Setup begins with a regional workspace in the OCI Console. Teams add provider connections and can map usage to people, managers, projects and cost centers. The service processes the data and sends it to a configured Fusion EPM instance, where teams review spending and plan budgets.

Person mappings can come from Fusion HCM or a CSV file. Project mappings use CSV files. Each workspace supports one destination. Processing covers the previous UTC day's data at 09:00 UTC, starting the day after onboarding, so this is a daily reporting workflow rather than a live per-request cost monitor.

That timing makes sense for a finance review. An engineering team investigating an unexpectedly expensive agent run will still need request-level traces and application context to explain what happened.

## Setup is more than connecting an API key

The [preparation guide](https://docs.oracle.com/en-us/iaas/Content/generative-ai/prepare-cost-advisor.htm) requires a workspace, at least one provider and a Fusion EPM destination. It also specifies an EPM Enterprise license and preparation of a Cost Advisor Planning application by an EPM administrator.

The documented onboarding includes EPM Automate on a customer-managed Windows computer or server, provider permission scopes, identity configuration and a dedicated integration user. This makes the product a more natural fit for enterprises already operating Oracle's planning tools than for a small team looking for an instant standalone spending dashboard.

Optional organizational mappings improve attribution, but someone still has to maintain them. A project transferred between teams can leave a misleading cost report if its ownership mapping does not change with it.

## What the savings recommendations cannot settle

Oracle's [product page](https://www.oracle.com/artificial-intelligence/cost-advisor-for-ai/) says additional insights can use optional prompt data, while core cost and usage processing does not require it. Customers retain control over decisions and subsequent actions. The announcement does not describe an automatic model-switching system.

Before acting on a suggested cheaper model, test it on the actual task. A model can cost less per token while producing more retries, more tool calls or more review work. For a document-processing workflow, the useful comparison is the cost of an accepted result at the same quality threshold.

**Cost visibility and workload optimization are separate jobs.** The first makes it easier to identify where to investigate. The second requires evidence about how a different configuration behaves.

## Availability and pricing limits

This is a limited-availability release, not general availability across OCI regions. Oracle's documentation is explicitly marked pre-GA and subject to change. The reviewed launch materials do not provide a public standalone price for Cost Advisor; the required EPM license and a team's existing provider contracts also need to be included in any evaluation.

The useful outcome would be a shared conversation between finance and engineering: which workflows justify their cost, which need a better configuration and which should stop. A consolidated bill is a starting point for that discussion, not its conclusion.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help enterprise AI teams explain technical products, operating costs and deployment tradeoffs clearly to buyers and decision-makers.

[Book a 30-minute call](https://cal.com/paras-madan/30min).