Microsoft Fabric is priced on capacity, not on data volume or headcount: you buy an F-SKU that grants a fixed pool of Capacity Units (CUs), and the right size is the one that absorbs your busiest hour without throttling. In US regions, an always-on F2 runs roughly $263 per month pay-as-you-go, an F64 roughly $8,410, and a one-year reservation cuts about 41% off any tier — but the number that decides your bill is which SKU you actually need, and that is set by refresh frequency and query concurrency, not by how many terabytes you store.

This guide is for the IT directors, data-platform owners, and finance leaders who have to size a Fabric capacity and defend the budget. It covers what a Fabric bill is actually made of, how CUs are metered, the reliable method for sizing an F-SKU, the F64 threshold that decides whether report viewers need paid licenses, and the levers that keep the number under control.

Key takeaways

• Capacity is the big number. A Fabric bill is compute (the F-SKU), plus OneLake storage billed separately at roughly $23 per TB per month, plus per-user Power BI licenses where they still apply.
• Data volume is the wrong sizing input. CUs are consumed by activity — refreshes, queries, pipelines, notebooks — so two organizations with identical data can land two SKU tiers apart.
• Size to your peak, then measure. The dependable method is to run real workloads on pay-as-you-go for two to four weeks, read the Fabric Capacity Metrics app, then reserve what the telemetry supports.
• F64 is the license cliff. At F64 and above, report viewers consume content for free; below F64, every viewer needs a paid Power BI license, which can flip the cheaper SKU into the more expensive decision.
• Reserve steady, pay-as-you-go for spiky. A one-year reservation saves roughly 41% for stable production; pay-as-you-go capacities can be paused, cutting a business-hours-only capacity to about a quarter of the always-on cost.

What does Microsoft Fabric cost in 2026? The short answer

Fabric bills on a single pool of compute measured in Capacity Units, so one capacity covers data engineering, warehousing, real-time intelligence, data science, and Power BI together. You buy an F-SKU, each tier providing double the CUs of the one below it, and you pay separately for OneLake storage and for any per-user Power BI licenses you still need, per the Microsoft Azure Fabric pricing page.

Three numbers frame the whole decision, using approximate US-region pricing as of mid-2026 (Solv Systems, Fabric pricing 2026):

• Smallest viable production platform: about $156 per month for an F2 reserved, plus paid licenses for whoever publishes content.
• Typical mid-market analytics estate: about $1,500 per month all in, on an F16 reserved with roughly 20 paid author licenses.
• Enterprise estate with a wide viewer audience: about $5,000 per month on F64 reserved, the point at which report viewers stop needing individual licenses.

What are you actually paying for in a Fabric bill?

Every Fabric bill has three components, and conflating them is where most budgets go wrong. Capacity is the dominant line item; the other two are easy to forget until they show up.

The three components of a Fabric bill

Capacity (the F-SKU)

• What it is: the shared pool of Capacity Units every Fabric workload draws from.
• How it is billed: pay-as-you-go by the hour, or a one-year reservation that saves roughly 41%.

OneLake storage

• What it is: the data you keep in Fabric's unified lake.
• How it is billed: separately, per GB per month, at roughly $23 per TB — cheap, but not zero and not part of the SKU.

Per-user Power BI licenses

• What it is: paid seats for people who publish or share Power BI content.
• How it is billed: about $14 per user per month for Power BI Pro; viewers are free at F64 and above.

How do Capacity Units, bursting, and smoothing actually work?

A Capacity Unit is Fabric's measure of compute, and your SKU number is simply CUs per second: an F2 gives you 2 CU/s, an F64 gives you 64. Every operation — an interactive report render, a scheduled refresh, a pipeline, a notebook — spends CUs out of that shared budget, and consumption is metered in short intervals rather than per job.

Two mechanisms keep short spikes from breaking your capacity, both documented in Microsoft Learn's capacity throttling guide. Bursting lets a single operation temporarily use more CUs than your SKU nominally provides so a heavy job finishes fast. Smoothing then spreads that borrowed compute across future time points so a brief peak does not immediately throttle you. Sustain the overspend for too long, though, and Fabric throttles — first delaying interactive requests, then rejecting operations — which is why sizing to your peak concurrency, not your average, is what actually protects the user experience.

Fabric does not slow down because your data is big; it slows down because too many people asked it to work at the same moment and the capacity had no headroom left.

How do you size the right Fabric F-SKU?

Start by discarding the instinct to size on data volume — that is how on-premises servers were bought, and it misleads here because OneLake storage is billed separately and cheaply. A ten-terabyte archive nobody queries consumes almost nothing; a fifty-gigabyte model refreshed every fifteen minutes and hit by 200 users all morning consumes a great deal (Solv Systems, Fabric capacity sizing).

Four factors drive CU consumption, in rough order of impact:

1. Refresh frequency and method. Full refreshes are expensive; incremental refresh is a fraction of the cost. Moving from four-hourly to hourly can add far more than 25% because each run carries fixed overhead.
2. Query concurrency. Not how many users exist, but how many query at the same moment. Concurrency peaks size your capacity; averages mislead.
3. Workload efficiency. Poorly written models and queries can push an F16 into throttling that a tuned workload would sail through on the same SKU.
4. Workload mix. Spark jobs, warehouse queries, and real-time ingestion all draw on the same pool, so the combination — not any single workload — sets the ceiling.

Because those factors are hard to model on paper, Microsoft's own P-to-F SKU right-sizing guidance and experienced practitioners converge on the same method: provision pay-as-you-go, run representative production workloads for two to four weeks, and read the telemetry before you commit to a reservation.

Sizing method compared

Size on data volume (the trap)

• Input used: terabytes stored.
• Result: systematically wrong — storage barely touches CU consumption.

Size on measured activity (the method)

• Input used: observed peak concurrency, refresh cadence, and workload efficiency from real runs.
• Result: a defensible SKU floor you can reserve with confidence.

How do you measure and right-size a live capacity?

The free Microsoft Fabric Capacity Metrics app is the instrument for this. It shows CU consumption, top consumers, throttling events, and whether smoothing is quietly saving you — the signals that tell you when to scale up or turn on autoscale. Its main limitation is retention: it keeps roughly 14 days of history, so you monitor continuously rather than reconstruct the past.

Once you can see the pattern, three levers move the number materially. First, pause and resume: pausing a pay-as-you-go capacity stops compute billing entirely, so a development or business-hours-only capacity can cost around a quarter to a third of the always-on figure. Second, right-size the SKU up or down based on measured peaks — but note that scaling across the boundary between F256 and F512 can briefly interrupt in-flight jobs, so do it in a maintenance window (Microsoft Learn, scale your Fabric capacity). Third, reserve once the pattern is stable to lock in the roughly 41% discount for predictable production load.

At BabyBots, this is exactly the work we scope as a fixed-fee engagement: we run a Microsoft Fabric capacity sizing and cost assessment that instruments your real workloads, reads the metrics, and hands back a defensible F-SKU floor and reservation plan — so you commit to a number your telemetry supports rather than a vendor estimate.

Pay-as-you-go or reserved: which should you buy?

The choice is about workload stability, not size. Pay-as-you-go bills hourly and can be paused, which suits evaluation, development, and spiky workloads. A one-year reservation is roughly 41% cheaper for the same SKU and suits steady production — the common break-even sits around 58% of monthly hours utilized, so anything running most of the time favors reserving.

Pay-as-you-go versus reserved

Pay-as-you-go

• Best for: evaluation, development, pilots, and bursty or seasonal workloads.
• Cost behavior: hourly billing you can pause; no commitment, higher per-hour rate.

Reserved (1-year)

• Best for: stable, always-on production capacity.
• Cost behavior: roughly 41% cheaper per SKU, committed for the term, cannot be paused away.

Approximate 2026 F-SKU price ladder (US regions)

F2 (2 CUs)

• Pay-as-you-go: about $263 per month always-on.
• Reserved (1-year): about $156 per month.

F16 (16 CUs)

• Pay-as-you-go: about $2,102 per month.
• Reserved (1-year): about $1,245 per month.

F64 (64 CUs)

• Pay-as-you-go: about $8,410 per month.
• Reserved (1-year): about $4,982 per month — and the tier where viewers stop needing licenses.

F128 and above (128–2,048 CUs)

• Pay-as-you-go: about $16,819 per month at F128, doubling with each step.
• Reserved (1-year): about $9,964 per month at F128, with the discount holding across the ladder.

Prices vary by Azure region and change over time, so confirm current rates on the Azure pricing page before you budget; the ratios between SKUs stay constant even when the absolute numbers move.

Frequently asked questions

How much does Microsoft Fabric cost per month?

For an always-on capacity in US regions, expect roughly $263 per month at F2, about $2,102 at F16, and about $8,410 at F64 on pay-as-you-go, with a one-year reservation cutting each by about 41%. On top of capacity you pay separately for OneLake storage at roughly $23 per TB per month and for any paid Power BI author licenses.

What F-SKU do I need?

The one that absorbs your peak concurrency and refresh load without throttling, which is rarely predictable from data volume. Run real workloads on pay-as-you-go for two to four weeks, read the Fabric Capacity Metrics app, and reserve the SKU your measured peaks support — two organizations with identical data often land two tiers apart.

What is the F64 threshold and why does it matter?

At F64 and above, users can view Power BI content without an individual paid license; below F64, every viewer needs a Power BI Pro license at about $14 per user per month. For a large viewer audience, that licensing math can make F64 cheaper overall than a smaller SKU once you add up the seats.

Is OneLake storage included in the capacity price?

No. OneLake storage is billed separately from the F-SKU at roughly $23 per TB per month. It is inexpensive relative to compute, but it is a distinct line item and should be budgeted on its own rather than assumed to be part of the capacity.

How can I reduce my Fabric bill?

The biggest lever is pausing pay-as-you-go capacity outside working hours, which stops compute billing and can cut a business-hours-only capacity to roughly a quarter of the always-on cost. Beyond that, right-size the SKU to measured peaks, tune inefficient refreshes and queries, and move stable production onto a one-year reservation for the roughly 41% discount.

Can I change my Fabric capacity after buying a reservation?

You can scale a capacity up or down at any time and pay the pay-as-you-go rate for the size you scale to, while scaling below your reserved level does not reduce the reservation you committed to. Be aware that resizing across the F256-to-F512 boundary can briefly interrupt running jobs, so schedule those changes during a maintenance window.

Where this is heading

Fabric is converging Microsoft's entire data and analytics stack onto one capacity model, which means capacity sizing is becoming a recurring FinOps discipline rather than a one-time purchase. As more AI and real-time workloads land on the same shared CU pool, the organizations that win are the ones instrumenting consumption from day one and treating right-sizing as an ongoing operating decision. If you are planning this alongside a move off legacy Power BI Premium, pair this guide with our Power BI Premium to Microsoft Fabric migration guide and the broader Fabric migration strategy overview, which sequence the timing and the estate work around the capacity decision.

Size your capacity before you commit

If you are staring at the F-SKU ladder trying to guess which tier you need, stop guessing and measure. Book a BabyBots Microsoft Fabric capacity sizing assessment — we instrument your real workloads on pay-as-you-go, read the metrics, and hand back a defensible F-SKU floor, a reservation plan, and the cost model to justify it, in a single fixed-fee engagement.

Sources

• Microsoft Azure — Microsoft Fabric pricing
• Microsoft Learn — Understand capacity throttling and smoothing
• Microsoft Learn — What is the Microsoft Fabric Capacity Metrics app?
• Microsoft Learn — Scale your Fabric capacity
• Microsoft Learn — Power BI Premium P SKU to Fabric F SKU migration decision guide
• Solv Systems — Microsoft Fabric pricing 2026 (F-SKU price list)
• Solv Systems — Microsoft Fabric capacity sizing