Capacity planning and add-ons

What it does

Dataverse capacity comes in three currencies — database, file and log — accrued from a tenant default plus per-licence entitlements, and consumed by every environment that isn't a trial, preview, support or developer environment. Add-ons top up whichever currency runs short, and process mining, AI and request capacity each have their own separate pools.

Key facts

Per-licence Dataverse accrual for Power Automate: Premium gives 250 MB database and 2 GB file per licence; Process and Hosted Process give 50 MB database and 200 MB file each.
Process mining data storage is 50 MB per Power Automate Premium licence, capped at 100 GB per tenant. The Process Mining add-on adds 100 GB of process mining capacity plus 2 GB Dataverse database and 1 TB Dataverse file.
Every environment consumes 1 GB whether or not it has a database. Trial, preview, support, developer and Teams environments don't count towards tenant capacity.
The default environment includes 3 GB database, 3 GB file and 1 GB log, and the list view only shows consumption above that included quota.
Database and file capacity are pooled across Dataverse and finance and operations workloads for enforcement. Log entitlement is tracked separately for Dataverse only.
Overflow works one way: spare database capacity can cover log and file overuse, spare log can cover file overuse. Spare file capacity covers nothing else.
Overage blocks admin operations — creating, copying, restoring or recovering an environment, converting a trial to paid, and adding a Dataverse database. Creating and recovering need at least 1 GB free.
Notifications fire weekly at less than 15% remaining, less than 5% remaining, and in overage. Admins can't opt out or delegate them.
Capacity can be preallocated to a specific environment, which reduces the tenant pool available to others. Environment-level allocation overage is currently soft-enforced by email at 85% of allocation.
Dataverse search indexes are reported as a DataverseSearch table and charged at the database rate. Turning search off deletes the index after 12 hours and breaks dependent generative AI experiences.

When to use / skip

Size capacity from the licence accrual first and buy add-ons only against measured consumption, because per-user accrual covers more than clients expect on a Premium-heavy tenant. Preallocating capacity to environments is worth doing when one team's growth would otherwise starve everyone else, and worth skipping on a small tenant where it just adds administration. The Process Mining add-on is only relevant where process mining is genuinely in scope — the seeded per-licence allowance is small and the add-on is tenant-wide.

Configuration decisions

Whether capacity is preallocated per environment or left pooled at tenant level.
Whether an environment can draw from the tenant pool when it exceeds its allocation, or is linked to a pay-as-you-go plan so overage bills to Azure instead.
Which environments are production or sandbox versus developer, since that decides whether they consume capacity at all.
Whether long-running audit logging is enabled, given log is the smallest entitlement and the one that overflows first.
Whether Dataverse search stays on, accepting the index storage cost, or off, accepting that dependent AI experiences degrade.
Whether daily capacity alerts are switched on per environment at a chosen threshold.

Gotchas

File overuse can't be covered by spare database or log capacity, so a tenant with hundreds of spare gigabytes elsewhere can still be in deficit and blocked from creating environments.
Attachments and annotations are progressively moving from database to file storage. Database usage falls, file usage rises, and it looks like a problem when it isn't.
An environment with preallocated capacity is measured against that allocation, not the tenant pool — allocating can create an overage that wouldn't otherwise exist.
Developer environments are free on capacity but expire after 90 days of no use, so they're not a place to park anything.
Indexes count towards database storage, and admin-configured Quick Find columns are a common and invisible cause of growth.
The Team Member licence gives no per-user database, file or log entitlement, so it won't appear in the entitlement breakdown.
Microsoft's Universal License Terms allow suspension of the service for sustained overage, with reasonable notice. It's not purely an alerting matter.

Consultant notes

Do the capacity sizing at the same time as the environment design. The number of production and sandbox environments is a capacity decision as much as an architecture one, and clients who design first and cost later end up deleting environments.
Push back on requests for a separate environment per project. Each one costs at least 1 GB before anything is built, and they accumulate faster than anyone deletes them.
Log capacity is the one that catches programmes out, usually six months after go-live when auditing has been on by default. Check it early.
Never quote add-on prices. Where Microsoft defers to the Power Platform Licensing Guide — and for capacity it does, repeatedly — refer the client there and to their account team.
Set up the capacity alerts during handover. Weekly threshold emails to an admin nobody reads are how a client discovers a problem at the point it blocks a release.

Worth revisiting if the file-versus-database overflow rules change, or after the next release wave adjusts per-licence accrual.