AWS Data Hub preview: what changes for FinOps
AWS Data Hub, announced November 20, 2026, lets the FinOps Agent query resource-level cost data with SQL. What changed, who it is for, and how to try it.
On November 20, 2026, AWS announced a preview of Data Hub in AWS Billing and Cost Management. It is a managed data layer that lets the AWS FinOps Agent answer resource-level cost questions with SQL, without you building a cost data export or pipeline. If your team asks "which resources drove this cost jump?" and the answer today means digging through the Cost and Usage Report (CUR), it is worth a look. If you already run a CUR pipeline into a warehouse, it does not replace it.
What was announced, and what changed?
AWS describes Data Hub as a managed data layer that the AWS FinOps Agent queries with SQL. According to the announcement, the dataset holds 13 months of hourly, resource-level cost and usage data in the CUR 2.0 schema, including resource identifiers and cost allocation tags. AWS provisions the data, refreshes it daily, and maintains the schema as new services and charge types appear.
The Data Hub documentation states the before and after plainly. Without Data Hub, the FinOps Agent answers at the service and account level using the AWS Cost Explorer API. With it, the agent can work with an individual resource, hourly detail, a resource identifier, a cost allocation tag or a usage quantity.
The announcement gives three example jobs for the agent:
- Identify the individual resources behind a cost change.
- Break down spend by team or cost allocation tag.
- Show how one resource's cost and usage change hour by hour.
- Billing datayour cost and usage, all regions
- Data Hub datasethourly, resource-level, CUR 2.0 schema, AWS-managed
- FinOps Agentqueries the dataset with SQL
- Your questionwhich resource, which team, which hour
Based on the AWS announcement and Billing and Cost Management documentation, November 2026.
Does it replace Data Exports?
No. The documentation says to keep using AWS Data Exports when you want to export cost data to Amazon S3, apply your own transformations, join it with data from outside AWS, or load it into a warehouse or BI tool. It also says Data Hub does not read your existing Data Exports and does not change them.
So the split is simple:
- Data Exports: you own the data, the pipeline and the joins.
- Data Hub: AWS holds the data, and only an agent queries it.
In the public preview, only the AWS FinOps Agent can use Data Hub. If your finance team needs the data in a spreadsheet or in your own dashboards, Data Hub does not give you that today.
Who should care, and who can ignore it?
Care if:
- Your AWS bill has grown and nobody can say which resources or teams are behind the change.
- Your tagging is decent, and you want to charge spend back to teams without maintaining a pipeline.
- You are a small team with no dedicated FinOps person, and a natural-language question is faster than building a query.
Ignore it for now if:
- You already load CUR data into a warehouse and your analysts are productive.
- You need the data outside AWS, or joined with revenue or customer data. That remains a Data Exports job.
- Your tags are missing or inconsistent. A resource-level query can only break spend down by a field that exists on the line item.
What does it cost, and what are the limits?
As of November 2026, AWS says there is no charge for Data Hub during the preview. The FinOps Agent's own product page does not state pricing and lists the agent as a preview, so do not assume the agent will stay free after preview. Check the pricing AWS publishes before you plan around it.
The documented limits and conditions:
- The preview is available in US East (N. Virginia), and the data covers cost and usage across all Regions.
- Only the AWS FinOps Agent can query it.
- A dataset created in the management account covers every account in the organization, including accounts that join later. A dataset created in a member account covers that account alone.
- The dataset uses the primary billing view of the account that creates it. If that account uses AWS Billing Conductor or billing transfer, the dataset holds pro forma costs, at the rates the account managing your billing sets. The preview has no billing view selector.
- Creating datasets needs Data Hub permissions, and a service control policy can deny
bcm-data-hub:CreateDataset.
How to try it safely this week
The prerequisites page describes two setups. You can create the dataset together with the agent in the FinOps Agent wizard, or create it first on the Data Hub page and connect an existing agent to it.
- Start with one account. Create a single-account agent and a dataset in a member account, so the data covers that account alone. This limits who can see what while you learn how it behaves.
- Pick a known question. Choose a cost change you already understand, such as a recent spike you traced by hand. Ask the agent and compare its answer with what you found.
- Check the numbers against Cost Explorer or your CUR data. Use a few resources and a few days. A mismatch is a finding, not a failure: it tells you how far to trust the agent.
- Review permissions. Decide who can create datasets and who can use the agent. Remember that a dataset created in the management account covers the whole organization.
- Move to the organization later. The documentation describes an organization-wide agent that runs in a delegated administrator member account, with the dataset created in the management account. Do that only after the single-account trial.
- 1Single-account dataset
- 2Ask a known question
- 3Compare with your bill data
- 4Review permissions
- 5Widen scope
Our suggested order, not an AWS procedure.
What we would check first
When we look at a preview feature like this, we start with the data, not the agent:
- Tag coverage. If, for example, a large share of your spend has no team tag, the answer to "which team?" will have a large unknown bucket. Fix tagging first, or at least measure it.
- Which cost view you are asking about. Pro forma and invoiced figures can differ. Know which one the dataset holds before you quote a number to finance.
- What a wrong answer costs. An agent answer is a lead to verify, not a ledger entry. Keep a human check before anyone acts on it, such as deleting a resource.
- Whether it changes a decision. If you will keep your CUR pipeline anyway, ask what question Data Hub answers faster than the pipeline does.
Preview features change. Re-read the documentation before you rely on this one, and treat the limits above as a snapshot of November 2026.
If you want a second opinion on cloud cost visibility, see our services.
General information about a vendor announcement, not financial advice. Check current AWS documentation and pricing before acting.