Aug 20, 2024Updated Sep 10, 2026

Power BI Embedded Pricing in 2026: What a Capacity Really Costs

ByPascal Kiefer

Power BI Embedded pricing confuses people for a good reason: the bill has three separate parts, Microsoft sells two different kinds of capacity for the same job, and the rule about who needs a license changes depending on how you embed.

This guide lays it out as it stands in 2026. It covers the F SKU and A SKU price points, how billing and pausing work, what reservations save, which people still need a Pro license, and three worked examples for a typical customer-facing portal. Prices are US list prices at the time of writing, rounded; they vary by region, so confirm your own numbers in the Azure pricing calculator before committing.

The three things you pay for

Capacity. A pool of compute in Azure that renders the reports. This is the big line item, and it is the same whether you serve ten viewers or ten thousand.
Per-user licenses for authors. The people who build and publish reports need Power BI Pro or Premium Per User. Since April 1, 2025 those cost $14 and $24 per user per month.
The application around the reports. Either engineering time to build it, or a portal product. Microsoft's API embeds a report; everything else is yours to provide.

Notably absent from the list: licenses for viewers. In the app-owns-data embedding model, Microsoft's documentation states that end users have no licensing requirements, on any capacity size. That single fact is why embedded analytics is cheaper than Power BI Pro the moment you have more than a few dozen external viewers.

F SKUs: Microsoft Fabric capacity

F SKUs are full Microsoft Fabric capacities bought through Azure. They run Power BI embedding plus every other Fabric workload, are the SKU Microsoft is investing in, and are the only option for Direct Lake semantic models. Pay-as-you-go pricing is about $0.18 per capacity unit per hour, which gives roughly these monthly figures for an always-on capacity (730 hours):

SKU	Capacity units	Pay-as-you-go, per month	1-year reservation, per month
F2	2	about $263	about $156
F4	4	about $526	about $313
F8	8	about $1,051	about $625
F16	16	about $2,102	about $1,251
F32	32	about $4,205	about $2,501
F64	64	about $8,410	about $5,003

US list prices, rounded, at the time of writing. A one-year reservation is roughly 41% below pay-as-you-go. Larger SKUs (F128 and up) scale linearly.

How F SKUs are billed

Per second, with a one-minute minimum, for as long as the capacity is running. There is no monthly commitment on pay-as-you-go.
Pausable. A paused capacity costs nothing. A portal used only during business hours can run the capacity for a third of the month.
Reservable. One- and three-year reservations lower the rate substantially, but you pay for the reserved capacity units every hour whether or not you use them, so a reservation and pausing do not combine. Reserve what runs around the clock; pay as you go for what doesn't.
Storage is separate. OneLake storage is billed per gigabyte per month and is not covered by a reservation. For a Power BI-only workload it is usually small.

A SKUs: Power BI Embedded capacity

A SKUs are the original Azure Power BI Embedded capacities. Despite the confusion caused by the Premium retirement, they are not being retired. They serve Power BI items only, no other Fabric workloads, and are billed hourly rather than per second, with no commitment and the same ability to pause. Approximate always-on monthly prices at US list rates:

SKU	Fabric equivalent	Per hour	Per month, always on
A1	F8	about $1.01	about $736
A2	F16	about $2.01	about $1,466
A3	F32	about $4.02	about $2,938
A4	F64	about $8.06	about $5,881

US list prices, rounded, at the time of writing. There is no A SKU smaller than A1, and no reservation pricing for A SKUs.

F or A for a new project?

Two things stand out from the tables. First, there is nothing below A1: if your workload is small, an F2 or F4 costs a third to two thirds of the smallest A SKU. Second, at equivalent sizes the A SKU is cheaper on pay-as-you-go than the F SKU (A1 versus F8, for instance), but the F SKU wins once you reserve it. The rule of thumb: small or intermittent workloads and anything touching Direct Lake or the wider Fabric platform go on F; a mid-sized, embedding-only workload that you don't want to reserve can still be cheapest on A. The Embedsy Portal treats the capacity as a setting, so switching later is a configuration change, not a rebuild.

Who still needs a per-user license

This is where most pricing estimates go wrong, in both directions. The rules as Microsoft documents them:

Report authors and workspace publishers: Pro ($14) or PPU ($24) per user per month, or a service principal for automated publishing. Every organization needs at least one.
Tenant and capacity admins: Pro or PPU.
Viewers in the app-owns-data model (embedded for your customers): none, regardless of capacity size.
Viewers in the user-owns-data model (embedded for your organization) or in the Power BI service itself: Pro or PPU on any A SKU and on F SKUs below F64. On F64 and larger, a Free license with the viewer role is enough.

So the widely repeated claim that you need an F64 to avoid viewer licenses is true for internal users in the Power BI service, and false for a customer portal built on app-owns-data embedding. If your viewers are customers or partners, an F2 serves them license-free just as an F64 would.

Three worked examples

Capacity, plus the Embedsy Portal on its Standard plan, which is $4 per portal user per month with 20 users included, starting at $80 per month. Azure resources for the portal itself (an App Service, a small SQL database and storage) are billed separately and are typically modest; the Azure documentation lists what gets deployed.

1. A customer portal used during business hours

Forty customers, a few concurrent viewers, reports refreshed overnight. An F2 handles this comfortably. With the portal's auto-pause, the capacity runs perhaps 11 hours on weekdays, about 240 hours a month: roughly $85 in capacity. Add the portal at $80 for the first 20 users and $4 for each of the other 20: about $245 per month all in. The same 40 viewers on Pro licenses would be $560 per month, and the portal comes with branding, RLS and scheduling that Pro doesn't.

2. An always-on partner portal with heavier models

Two hundred partner users across time zones, larger semantic models, a few dozen concurrent sessions at peak. An F8 is a sensible starting size, and because it runs around the clock a one-year reservation makes sense: about $625 per month. Portal at $80 plus 180 additional users: $800. Around $1,425 per month, versus $2,800 per month for 200 Pro licenses, before considering that Pro doesn't give partners their own branded environment.

3. Heavy internal and external use on one capacity

If you also want thousands of internal employees to consume content in the Power BI service without Pro licenses, only F64 or larger does that. A reserved F64 is about $5,000 per month and covers both the internal audience and an embedded customer portal on the same capacity. Below roughly 350 internal Pro users this is more expensive than the licenses; above it, it isn't.

How to keep the capacity bill down

Pause when nobody is there. The Embedsy Portal starts the capacity when a user logs in and pauses it when the portal is idle, and lets you schedule on and off times so overnight refreshes still run. For a business-hours workload this alone cuts the pay-as-you-go bill by more than half.
Right-size before you upgrade. Most "we need a bigger SKU" conversations are really semantic model problems. Our post on right-sizing Fabric capacity explains how to tell, and our capacity optimisation service does the work if you'd rather not.
Reserve only the baseline. Reserve the capacity that runs 24/7 and cover peaks with pay-as-you-go scaling rather than reserving for the peak.
Set an Azure budget. Alerts at 50%, 80% and 100% of your expected spend catch a capacity that was left running. The Azure budget guide in our documentation walks through it.

Key takeaway

Power BI Embedded pricing comes down to one capacity, a handful of Pro licenses for authors, and whatever you use to build the portal. Viewers cost nothing in the app-owns-data model, on any capacity size, which is what makes the economics work for customer-facing analytics. Start small on an F2 or F4, pause it when it's idle, and reserve only once you know what runs all day.

If you'd like a second opinion on sizing for your workload, that's a conversation we have every week.

Book a free demo or start the free trial