Fabric SKU comparison: F2 to F2048

Prices and Microsoft-published limits for every Fabric capacity, in one table. The two numbers that usually decide the size are the memory per semantic model (your largest model must refresh inside it) and whether you need free viewers (F64 and up).

All SKUs at a glance

US East list prices; limits per Microsoft Learn. Verified 2026-10-10.
SKUCUsPower BI equiv.Memory / modelDirect Lake rows / tablePAYG / monthReserved / monthFree viewers
F22—3 GB300 million$263$156No
F44—3 GB300 million$526$313No
F88EM1/A13 GB300 million$1,051$625No
F1616EM2/A25 GB300 million$2,102$1,251No
F3232EM3/A310 GB300 million$4,205$2,501No
F6464P125 GB1.5 billion$8,410$5,003Yes
F128128P250 GB3 billion$16,819$10,005Yes
F256256P3100 GB6 billion$33,638$20,011Yes
F512512P4200 GB12 billion$67,277$40,021Yes
F10241,024P5400 GB24 billion$134,554$80,043Yes
F20482,048—400 GB24 billion$269,107$160,085Yes

How to choose

  1. Size for the largest model, not the average. A full Import refresh needs roughly twice the model's size in memory, and it must fit inside the per-model limit.
  2. Decide on viewer licensing. If you have hundreds of viewers, F64's free-viewer benefit can outweigh the capacity cost; below F64 the capacity is an addition to Pro licences.
  3. Check Direct Lake guardrails. Rows and Parquet files per table are capped per SKU; exceeding them causes DirectQuery fallback or failed refreshes.
  4. Measure before you scale. Throttling is often caused by colliding refreshes or inefficient DAX that a bigger SKU only postpones.

Frequently asked questions

Which Fabric SKU replaces Power BI Premium P1?
F64 is the same capacity size as P1 (64 capacity units, 8 Power BI v-cores) and keeps the free-viewer benefit. Many teams can run smaller once models are optimised, but anything below F64 requires every viewer to hold a Pro or PPU licence.
What is the cheapest way to run Fabric capacity?
A 1-year reservation is about 40.5% cheaper than pay-as-you-go at list price. Pay-as-you-go capacity can be paused, which can be cheaper for dev/test capacities that are idle most of the day.
Does a bigger SKU make a slow report faster?
Only if the capacity is actually saturated. Slow visuals are most often caused by model design and DAX — high-cardinality columns, bidirectional relationships, or measures that force row-by-row evaluation. Check the Capacity Metrics app and the model before scaling up.
Where does the free-viewer benefit start?
At F64. On F64 and larger, users with a free Fabric licence and the Viewer role can consume Power BI content. On F2 to F32 every viewer needs Pro or PPU.

Sources: capacity limits, Direct Lake guardrails, Fabric pricing, Power BI licensing.