The Basics

If you’ve dealt with Power BI licensing before, Fabric licensing makes sense as an extension of that model plus some new parts around CUs, bursting and smoothing. But what if you are brand new to Fabric, Power BI, and possibly even Office 365?

If you want to get started with Fabric, you need at a bare minimum the following:

1. Fabric computing capacity. The cheapest option, F2, costs $263 per month for pausable capacity (called Pay-as-you-go) and $156 per month for reserved capacity. Like Azure, prices vary per region.
2. An Entra tenant. Formerly called Azure Active Directory, Entra is required for managing users and authentication.
3. Fabric Free license. Even though you are paying for compute capacity, all users need some sort of license applied to them as well. I think assigning a license requires an office 365 tenant to access the admin portal but I’m not sure.

Once you have an F2, you can assign that capacity to Fabric workspaces. Workspaces are basically fancy content folders with some security on top of it. Workspaces are the most common way access is provided to content. With the F2 you’ll have access to all non-Power BIfabric objects.

The F2 sku provides 0.25 virtual cores for Power BI workloads, 4 virtual cores for Spark workloads, and 1 core for data warehouse workloads. These all correspond to 2 CUs, also known as compute units. CUs are a made up unit like DTUs for databases or Fahrenheit in America. They are, however, the way that you track and manage everything in your capacity and keep costs under control.

Storage is paid for separately. OneLake storage costs $0.023 per GB per month. You also get X TB of free mirroring storage equal to your SKU level. So F2 gets 2 TB of storage.

There is no cost for networking, but that will change at some point in the future.

Power BI content

If your users want to create Power BI reports in these workspaces, they will need to be assigned a Power BI Pro license at a minimum, which costs $14 per user per month. This applies to both report creators and report consumers. Pro provides a majority of Power BI features.

The features this does not provide are covered by Power BI Premium per User (PPU) licenses, which cost $24 per user per month. These licenses allow for things like more frequent refreshes and larger data models. PPU is a hybrid license because you both license the user as well as assign the content to a workspace set to PPU capacity.

One of the downsides of the PPU model is that they act as a universal receiver of content but not a universal donor. Essentially, the only way for anyone to read reports hosted in a PPU workspace is to have a PPU license. So, you can’t use this as a cheat code to license your report creators with PPU and everyone else with Pro. Nice try.

There is demand for a fabric equivalent, a FPU license, but there is no word on when or if this will happen. Folks estimate this could cost anywhere from $30 to $70 per user per month if we get one.

Finally, if you ramp up to an F64 sku, Power BI content is then included. Users will still need a Fabric Free license. At $5002/mo for F64, this means it’s worth switching over at 358 Pro users or 209 PPU users. Additionally, you unlock all premium features including copilot.

Even if you pay for F64 or higher (or Power BI report server on Prem), any report creators need to be licensed with Power BI Pro for use of that publish button. I cannot understand why Microsoft would charge $5k per month and then charge for publishing on top.

There are also licensing complications for embedding Power BI in a custom application which is outside of the scope of this post.

Capacity management

Despite a Fabric SKU providing a fixed number of Capacity Units, Fabric is also intended to be somewhat flexible. Fabric customers like the pricing predictability of Fabric compared to Azure workloads, but because of the sheer number of workloads supported, actual usage can vary wildly compared to when premium capacity was only Power BI reports.

In order to support that, Fabric allows for bursting and smoothing. This is similar to auto-scaling, but not quite. Bursting will provide you with more capacity temporarily during spikey workloads, by up to a factor of 12 in most cases. However this bursting isn’t free. You are borrowing against future compute capacity. This means it’s possible to throttle yourself.

Bursting is balanced out by smoothing. Whenever you have exceeded your default capacity, future work is spread out over a smoothing window. This is a 5 minute window for anything a user might see and 24 hours for background tasks. If you are using pay-as-you-go capacity, you’ll see a spike in CUs when you shut down the capacity as all of this burst debt is paid off all at once instead of waiting for smoothing to catch up.

From what I’ve been told by peers, it’s possible that you can effectively take down a capacity with a rogue Spark notebook by bursting for so long that smoothing has to use the full window to catch up. At Ignite they announced they are working on Surge protection to prevent this

Capacity consumption can be monitored with the Fabric Capacity Metrics App.

I believe you can also upgrade a reserved capacity temporarily and pay the pay-go costs for the difference, but I can’t find docs to that effect.