Generative AI projects tend to start the same way in most organizations.

A team builds a chatbot. Another experiments with document summarization. Someone launches an internal assistant for customer support or engineering. None of these projects look particularly large on their own.

Then the Bedrock bill arrives.

Usage has spread across teams, environments, and prototypes. Costs have grown quickly. And suddenly someone in finance asks the obvious question: where exactly is this spending coming from?

Until recently, Amazon Bedrock made that question harder to answer than it should have been.

When GenAI costs become a blind spot

Before AWS introduced the Projects API, all Bedrock inference usage within an AWS account appeared as a single aggregated cost in Cost Explorer.

From a financial perspective, everything was mixed together. A production application and an experimental prototype generated the same type of cost line. Identifying which team or product was responsible required additional tracking mechanisms or custom instrumentation.

For organizations trying to apply FinOps practices to AI workloads, this created a growing blind spot. Generative AI adoption was accelerating, but the visibility required to manage that spending was missing.

A structural change inside Bedrock

AWS recently introduced the Projects API in Amazon Bedrock, allowing organizations to organize GenAI workloads into logical projects.

Each project can have its own IAM permissions and cost allocation tags. In practical terms, this allows companies to structure their AI workloads around teams, products, or environments.

Instead of one undifferentiated Bedrock cost bucket, workloads can now be separated and associated with specific initiatives.

For FinOps teams, this is more than a technical detail. It makes it possible to connect AI spending to the teams and applications that generate it.

Why this matters now

Generative AI has rapidly become one of the fastest-growing sources of cloud spending. Model inference, prompt experimentation, and production workloads can generate significant usage in a short amount of time.

Without clear attribution, it becomes difficult to answer basic operational questions. Which teams are generating the most inference traffic? Are premium models being used where smaller models would suffice? Which initiatives are actually delivering value?

Project-level isolation helps bring AI workloads back into the financial governance framework that already exists for the rest of the cloud.

What sits behind the feature

Under the hood, Bedrock projects rely on Mantle, which exposes APIs compatible with OpenAI, including the Responses API and Chat Completions API.

For teams already building applications with OpenAI-style integrations, the transition to Bedrock can often be relatively smooth while providing stronger governance around costs and permissions.

A deeper analysis from the Stable team

The introduction of Bedrock projects is an important step toward making generative AI spending easier to track and manage.

The Stable team recently published a deeper technical analysis of the Projects API and its implications for cost attribution and AI workload governance.

You can read the full article here: https://www.stableapp.cloud/blog/amazon-bedrock-projects-api-finally-assigning-genai-costs-by-team-and-project/