AI cost control becomes mandatory for enterprise

OpenAI introduced new usage analytics and updated spend controls for ChatGPT Enterprise on June 18. At almost the same time, Microsoft describes a consumption-based model for the Work IQ API through Copilot Credits. For enterprises, the message is clear: AI consumption is becoming an operating and finance topic, not just an IT feature.

What is changing

According to OpenAI, the Global Admin Console now brings ChatGPT and Codex credit usage into one view. Administrators can track trends over time, identify top users, and break down consumption by workspace, user, product, and model. OpenAI also points to a unified Cost API, so the same usage data can be pulled into internal systems.

For limits, OpenAI is moving beyond one-size-fits-all caps. Workspace defaults, group limits, and individual overrides can be combined. According to the Help Center, users can see their limits and request increases with context when they need more capacity.

Microsoft applies a similar control logic to the Work IQ API: custom agents in Copilot Studio, Foundry, or third-party platforms can consume Copilot Credits when they use Microsoft 365 data through Work IQ. Before production use, administrators are expected to configure billing, access, limits, and alerts.

Why CIOs and CFOs should care

This creates a new control layer between platform teams, business units, and finance. An agent is not just an application with a license fee. It generates ongoing model, context, and tool-call costs — often exactly where many small tasks are automated.

For regulated organizations, this is also governance: Who may use which models? Which department creates which consumption? When does a power user need more budget? And which agents produce enough value to justify their credits?

What to review now

Start with a simple AI FinOps model: cost centers, technical limits, alerts, review points, and clear owners for each workspace or agent. Only then should the rollout become broader.

The key question is no longer: “Do we have access to AI?” It is: “Can we run AI in a way that keeps value, risk, and consumption visible?”