Power BI

Overview

Power BI is now a workload inside Microsoft Fabric rather than a standalone service, which is the structural fact that dates most older descriptions and moves its capacity documentation and billing into the Fabric estate.

It comes in three parts: a Windows authoring application, the hosted service, and gateways for reaching data on your own network.

Its engine is an in-memory columnar store, and the modelling choice is yours: import data into it, query the source directly, mix the two, or read lakehouse tables in place.

The licensing trap is the one to understand first, because it decides most budgets. Below a particular capacity size every reader still needs a paid per-user licence; at or above it, readers need none.

Prices have moved too. The two per-user tiers are now fourteen and twenty-four dollars a month, so figures of ten and twenty are out of date.

Capacity is bought as Fabric units, and the older premium capacity names are legacy, which is worth knowing when reading anything written before the merge.

Its assistant is gated more tightly than people expect: it needs paid capacity at a small but non-zero size, a per-user licence alone is not enough, and it is not supported on sovereign clouds.

Naming keeps moving around it. The agent feature was renamed, and a semantic and ontology layer is appearing above the model in the wider Fabric metering.

Governance is one of its genuine strengths, with row and object-level security enforced even through the assistant, sensitivity labels, endorsement, lineage and deployment pipelines.

Its reach is the other strength: hundreds of connectors, native lakehouse integration, and an analysis endpoint that lets other tools query the model directly.

Key features and capabilities

The same headings are used for every bi & dashboards entry, so two tools can be read side by side.

Data modelling and semantic layer

• A tabular semantic model authored in the desktop application, with its own calculation language for measures and columns
• Four connection modes, importing into the in-memory engine, querying the source live, mixing both, or reading lakehouse tables directly
• Relationships carry cardinality and filter direction, and a separate query language handles transformation
• Per-capacity memory limits cap model size, and can force a fallback from direct lakehouse reads to live queries
• A semantic and ontology layer is emerging above the model in the wider platform's metering

Charts, dashboards and interactivity

• A large built-in visual set plus third-party and custom visuals from the marketplace
• Recent releases moved to a modern default visual style with a theme pane and new slicer and matrix behaviour
• Cross-filtering and drill-through are native rather than configured
• Dedicated phone layouts in the mobile application
• Lakehouse file locations can now be used directly as image sources

Calculations, AI and natural language

• Its own calculation language is the analytical surface, alongside a natural-language question visual
• Assistant-generated summaries and narrative visuals, which respect row and object-level security
• Data-driven alerts and an event-based alerting service
• Agents provide the newer conversational layer, renamed recently, and are gated behind paid capacity
• The assistant requires paid capacity at a small size; a per-user licence alone is not sufficient

Publishing, embedding and collaboration

• Publish to workspaces and organisational applications, with email subscriptions and scheduled delivery
• Embedding through a dedicated service and client interfaces for white-labelled use
• Comments on report pages and visuals now work inside organisational applications, with mentions
• Export to document, presentation, spreadsheet and delimited formats, plus paginated report formats
• Readers need no licence only at or above the larger capacity threshold

Security, lineage and certification

• Row-level and object-level security, both enforced through the assistant as well as the report
• Sensitivity labels, and endorsement of content as promoted or certified
• Lineage and impact analysis through the wider governance service
• Deployment pipelines for development, test and production promotion
• A capacity metrics application for monitoring usage and throttling

Integrations

• Hundreds of connectors through its transformation query layer
• Native integration with the platform's own lakehouse and warehouse
• Transformation, workflow and legacy integration jobs now appear as metered items in the same platform
• Published interfaces, an analysis endpoint for external query tools, and embedding toolkits
• A software development kit for building custom visuals

Where it runs and what it costs

• The service is software as a service only; a separate on-premises report server product exists but is not covered here
• Gateways bridge to data on your own network or in a private network
• Priced two ways, per user or by capacity, and the capacity threshold decides whether readers need licences
• The assistant is not supported on sovereign clouds because of accelerator availability
• Capacity can be paid as you go or reserved for one or three years

Pricing

SubscriptionFree account; $14 per user per month

Two models that interact, and the interaction is where budgets go wrong. Per user, the professional tier is $14 and the premium per-user tier $24 a month on annual billing, with a free account for personal use. Capacity is bought in platform units at about eighteen cents per unit-hour pay-as-you-go, three times that for overage, or roughly $938 per unit on a one-year reservation, which is about a forty per cent saving.

Vendor pricing page →

Demos and videos

• Monthly product update Webinar or talk · YouTube
• Direct lakehouse reads explained Tutorial · YouTube

About Microsoft

Power BI is built and operated by Microsoft as part of the Fabric platform, so its commercial terms follow that product rather than standing alone. For the financial year ended June 2026 Microsoft reported revenue of $331.8bn, up eighteen per cent, with its cloud segment at $59.3bn in the final quarter, though none of those figures is specific to this product and no customer count for it is published. As with the other hyperscaler entries in this catalogue, corporate details such as employee numbers are not published on the documentation domains that carry the product detail. Two naming changes are worth carrying: premium capacity is now bought as Fabric capacity units, and the agent feature has been renamed, so older licensing guides describe a structure that no longer exists.

Redmond, Washington · microsoft.com