Power BI Embedded Pricing: A SKUs vs Fabric F SKUs vs Per-User Licences
What it costs to show Power BI reports inside your own app, which SKU fits which embedding scenario, and when an F SKU is cheaper than an A SKU.
Core Engineering Takeaways
- •There are two embedding scenarios: embed for your customers (your app owns the data) and embed for your organisation (users sign in with their own accounts). They have different licence rules.
- •When embedding for customers, end users need no Power BI licence; you pay for capacity (an A SKU or an F SKU).
- •A SKUs start at $736/month for A1. The A4 ($5,881/month) is the same size as an F64, which costs $5,003/month on a one-year reservation.
- •A SKUs suit pure embedding with variable load; F SKUs suit teams that also use Fabric or want reservation pricing.
1. First decide which kind of embedding you are doing
"Power BI Embedded" covers two different setups, and the licensing rules are completely different.
- Embed for your customers (app owns data). Reports appear inside your product or portal. Your users never sign in to Power BI; your app authenticates with a service principal and generates embed tokens. End users need no Power BI licence — you pay for capacity.
- Embed for your organisation (user owns data). Reports appear in an internal app such as SharePoint, Teams or an intranet, and users sign in with their own Microsoft Entra accounts. Each user needs a Pro or PPU licence, unless the content is on a capacity large enough for free viewers.
If your users are customers outside your organisation, you are almost certainly in the first scenario, and this guide's capacity pricing applies.
2. A SKU prices and their F SKU equivalents
Microsoft's embedded capacity documentation maps each A SKU to an F SKU of the same size. US East list prices, checked on 2026-10-10:
- A1: 1 v-core, 3 GB RAM, $736/month. Same size as F8, which is $1,051 pay-as-you-go or $625 reserved
- A2: 2 v-cores, 5 GB RAM, $1,466/month. Same size as F16, which is $2,102 pay-as-you-go or $1,251 reserved
- A3: 4 v-cores, 10 GB RAM, $2,938/month. Same size as F32, which is $4,205 pay-as-you-go or $2,501 reserved
- A4: 8 v-cores, 25 GB RAM, $5,881/month. Same size as F64, which is $8,410 pay-as-you-go or $5,003 reserved
- A5: 16 v-cores, 50 GB RAM, $11,768/month. Same size as F128, which is $16,819 pay-as-you-go or $10,005 reserved
- A6: 32 v-cores, 100 GB RAM, $23,543/month. Same size as F256, which is $33,638 pay-as-you-go or $20,011 reserved
A7 and A8 exist for very large deployments but are not priced on the public page; both A and F SKUs support embedding for customers and for your organisation.
Confirm current prices on the Power BI Embedded pricing page.
3. A SKU or F SKU?
Size for size, the A SKU is cheaper than an F SKU on pay-as-you-go, and the F SKU is cheaper once you reserve it. The right choice depends on your load pattern and what else you run.
Choose an A SKU when
- You only embed Power BI and do not need lakehouses, notebooks or other Fabric items.
- Usage is concentrated in business hours or in particular regions, so you can pause or scale the capacity and pay only while it runs.
- You want the capacity managed purely in Azure alongside the rest of your application.
Choose an F SKU when
- You want the one-year reservation discount on a capacity that runs all the time.
- You also use Fabric workloads (Direct Lake, lakehouses, pipelines) or Copilot.
- The same capacity also serves internal users, who can view with free licences from F64.
4. How to size an embedded capacity
Capacity cost is driven by the work your reports generate, not by the number of customers. The main levers:
- Model memory. The largest model must fit the SKU's per-model memory limit, including refresh overhead. A model that needs more than an A1/F8 allows forces a larger SKU regardless of traffic.
- Concurrency at peak. Many users opening the same report at 9am cost far more than the same users spread across the day.
- Query efficiency. One slow measure multiplied by thousands of customer sessions is the most common reason embedded capacities are oversized. Fixing the measure is cheaper than the next SKU.
- Refresh schedule. Refreshes use the same capacity as queries; schedule them away from peak usage.
Optimise the model first, then load-test, then pick the SKU — in that order.
5. Common embedded licensing mistakes
- Using a master user account in production. Use a service principal; it is what Microsoft recommends for app-owns-data embedding.
- Giving every customer a Pro licence. For external customers in an app-owns-data setup, capacity replaces per-user licences.
- Running the capacity 24/7 when customers use it 10 hours a day. On an A SKU or a pay-as-you-go F SKU, schedule pause and resume.
- Ignoring RLS. Multi-tenant embedding needs row-level security or separate models per customer — see the row-level security guide.
Sizing an embedded capacity?
Estimate the SKU your models and refresh schedule need, or have me review the model before you pay for a bigger capacity.
Get the next Engineering Note
Practical DAX, semantic model and Fabric notes. 1–2 a month, unsubscribe any time.