Microsoft Fabric Pricing: Is It Worth It In 2026? [Reviewed]

Microsoft Fabric is billed by capacity: you rent capacity units (CUs) on Azure at $0.18 per CU-hour pay-as-you-go (US West 2 list price, October 2026), so the smallest F2 costs $262.80 a month and an F64 costs $8,409.60 a month, or about $5,003 a month on a one-year reservation. OneLake storage is billed on top, at about $0.023 per GB a month. Below F64, every Power BI report viewer also needs a Pro license ($14 per user a month) or Premium Per User ($24 per user a month), both billed annually; from F64 up, viewers with a free license can read reports. Copilot needs a paid F2 or larger (or a P1) capacity; Pro, PPU or a trial capacity on its own doesn't include it.

Microsoft Fabric capacity pricing for every F SKU

The number in an F SKU's name is its CU count: an F2 has 2, an F64 has 64. At Microsoft's reference price of $0.18 per CU per hour, a SKU costs its CU count × $0.18 an hour, and 730 hours make a month.

The one-year reservation is $938 per CU a year, about 40.5% below running the same CU all year on pay-as-you-go ($0.18 × 8,760 hours = $1,576.80). Microsoft's Fabric pricing page rounds that to "save ~41%". The $938 rate comes from the 2026 price breakdowns published by EPC Group and AlphaBOLD; the reserved monthly figures below are $938 × CUs ÷ 12.

US West 2 list prices in USD as of October 2026, before storage, Power BI licenses and tax; other regions differ, so check the Azure pricing calculator for yours. The P equivalents come from Microsoft's P SKU to F SKU migration guide, which treats them as compute equivalents only.

• Per second, not per month. Pay-as-you-go is charged for every second the capacity is running, with a one-minute minimum, and can be resized any time with no commitment; the monthly figure assumes you never pause.
• A reservation is a block of CUs, not a capacity. You buy a number of CUs for one Azure region, and Microsoft applies the discount hour by hour to whatever F capacities run there. Per Microsoft's reservation docs, reserved CUs with no matching usage in an hour are lost (so pausing a reserved capacity saves nothing), the reservation covers compute only, and it doesn't renew on its own: when it expires, the capacity keeps running at pay-as-you-go rates.

Pay-as-you-go or reservation?

The reservation wins once a capacity runs more than about 59.5% of the hours in a year: $938 ÷ $0.18 = 5,211 hours, out of 8,760, or about 14.3 hours a day. A capacity that serves Power BI reports has to stay on, so it almost always belongs on a reservation. A capacity used only for data engineering or development is different: an F64 running 12 hours on weekdays is 3,120 hours a year, so pay-as-you-go comes to 64 × $0.18 × 3,120 = $35,942.40 a year, against $60,032 for a reserved F64 (64 × $938).

Pausing and resuming Fabric capacity

Pay-as-you-go capacity can be paused and resumed from the Azure portal, and compute billing stops while it's paused. Two things catch people out:

• Pausing settles your debts. Any smoothed usage still waiting to be counted (next section) and any overage is billed at once, which shows up as a spike in the Capacity Metrics app.
• Storage keeps billing. OneLake data on a paused capacity is still charged per GB, and its reports and items are unavailable until you resume.

The common pattern is one reserved capacity for reports people read during the day, plus a pay-as-you-go capacity for heavy or occasional work that gets paused when it's idle.

Smoothing, bursting and throttling in plain words

A capacity gives you a fixed amount of compute per second: an F64 has 64 CU-seconds every second, or 5,529,600 a day. Per Microsoft's throttling and smoothing docs:

• Bursting: a heavy job can briefly use more compute than your SKU provides, so a big refresh finishes instead of queueing.
• Smoothing: Fabric then spreads the cost of that work over time. Interactive operations (report clicks, queries) are spread over at least 5 minutes; background operations (refreshes, Spark and SQL jobs, Copilot) over 24 hours. This is why you can size for your average load rather than your worst minute.
• Throttling: if smoothed usage keeps running above what you bought, you are borrowing from the future. Up to 10 minutes of future capacity, nothing happens. From 10 to 60 minutes, interactive requests are delayed by 20 seconds. From 60 minutes to 24 hours, interactive requests are rejected while background jobs still run. Past 24 hours, everything is rejected until the backlog burns down.

In practice it shows up as slow or failing reports the morning after a heavy overnight load. The fixes: move work off peak, split workloads across capacities, or buy a bigger SKU.

Two newer options change that trade-off. Capacity overage, introduced in preview in April 2026 for F SKUs, lets an admin pay for excess usage instead of being throttled: the excess is billed at three times the pay-as-you-go rate, up to a rolling 24-hour limit in CU-hours. Microsoft suggests keeping that limit under a third of your daily CU-hours, because beyond that the next SKU up costs about the same. And at FabCon Europe in September 2026, Microsoft announced a zero-provisioned F0 SKU with on-demand billing, listed as coming soon in preview. As of October 2026 its prices aren't published, so I'd leave it out of any budget for now.

OneLake storage pricing

Storage is the second meter, and it doesn't use your CUs. Per Microsoft's OneLake consumption docs, it's billed per GB per month on pay-as-you-go:

• Hot storage: about $0.023 per GB a month (roughly $23 per TB) in US West 2 and Central US, and $0.026 in East US, per the Azure retail price list as reported by Dawiso in October 2026. Cheaper cool and cold tiers are on the pricing page as well.
• Disaster recovery: with BCDR on, the data is geo-replicated and billed at the higher BCDR rate (about $0.0414 per GB a month for hot data).
• Power BI import models are covered by the Power BI license, not billed as OneLake storage.
• Mirroring has a free allowance: 1 TB of mirrored storage per CU, so 64 TB on an F64, per the mirroring docs. You pay beyond that, or while the capacity is paused.
• Soft-deleted data is billed like active data until it's purged.

For most BI teams, storage is small next to compute: 2,000 GB of hot data is about $46 a month in US West 2 (2,000 × $0.023).

Fabric trial capacity

The Fabric trial is free for 60 days and runs as an F4 or, if you're eligible, an F64-sized capacity. When it ends, you have seven days to move non-Power BI items to a paid capacity before they're gone. It's fine for testing pipelines, lakehouses and report performance, but not Copilot in Power BI: Microsoft's Copilot setup docs say trial capacities aren't supported. For that, a paid F2 for 14 days around the clock is 336 hours × $0.36 = $120.96.

Power BI licenses on Fabric: when viewers need Pro

Buying capacity removes Power BI licenses only for viewers, and only from F64 up. Per Microsoft's Fabric licensing docs, with list prices from the Power BI pricing page:

• Authors and editors need Pro ($14 per user a month) or Premium Per User ($24 per user a month), both billed annually, to create or change content in a shared workspace, whatever the capacity size.
• Viewers on F2 to F32 each need Pro, PPU or an individual trial.
• Viewers on F64 and larger can use a free Fabric license, as long as they have the Viewer role on the workspace. External guests follow the same rule.
• Premium Per User is not a capacity. It can't run non-Power BI Fabric items, and content in a PPU workspace, including access over the XMLA endpoint, needs a PPU license for every person who opens it (PPU FAQ).

For Power BI's own plans (Free, Pro, Premium Per User, Embedded), see our Power BI pricing guide.

What 25, 100 and 500 viewers cost on Fabric

Per month, with five authors on Pro, everyone else only reading reports, and US list prices (the calculator at the top runs this for your own team):

At 500 viewers the reserved F64 saves $1,997.33 a month, or $23,968 a year. A reserved F64 beats Pro seats from 358 viewers ($5,002.67 ÷ $14 = 357.3), and a pay-as-you-go F64 from 601 viewers ($8,409.60 ÷ $14 = 600.7).

The math changes if you need capacity anyway for Fabric workloads such as lakehouses, pipelines or notebooks. On an F32, viewers still need Pro, and stepping up from a reserved F32 ($2,501.33) to a reserved F64 ($5,002.67) costs another $2,501.33 a month, which is 179 Pro seats ($2,501.33 ÷ $14 = 178.7). So a team already running Fabric should move reports to F64 from about 179 viewers, not 358.

What Copilot in Fabric and Power BI requires (October 2026)

Copilot has no per-user price: it runs on a capacity and spends that capacity's CUs. As of October 2026, per Microsoft's docs:

• SKU: a paid F2 or larger, or a Power BI Premium P1 or larger (Copilot in Fabric setup).
• Pro and PPU workspaces: not supported directly. An admin can designate an F2+ or P1+ capacity as a Fabric Copilot capacity, which then pays for a chosen group's Copilot usage wherever their content lives. It has to be in the tenant's home region, and capacities in Embedded license mode can't be used.
• Tenant settings: "Users can use Copilot and other features powered by Azure OpenAI" must be on. Microsoft's Copilot admin settings now list it as on by default, and admins can limit it to security groups.
• Regions: if your capacity sits outside the US and the EU Data Boundary, Copilot stays off until an admin allows "Data sent to Azure OpenAI can be processed outside your capacity's geographic region" (Copilot overview). Microsoft's own pages disagree on that setting's default, so check it in the admin portal.
• Fabric data agents have the same capacity requirement and their own tenant settings (data agent docs).

What it costs per question: Microsoft's Copilot consumption page charges 100 CU-seconds per 1,000 input tokens, 10 per 1,000 cached input tokens and 400 per 1,000 output tokens, as a background operation smoothed over 24 hours. Its own example, 2,000 input tokens and 500 output tokens, comes to 400 CU-seconds: about two cents at $0.18 per CU-hour ($0.18 × 400 ÷ 3,600 = $0.02). An F2 has 172,800 CU-seconds a day, so 432 requests of that size would use a whole day of an F2 with nothing else running. A real conversation usually takes several requests, so treat that as a ceiling, not a forecast.

The cheapest route to Copilot for the 25-viewer team above is a pay-as-you-go F2 as its Copilot capacity: $420 in Pro seats + $262.80 = $682.80 a month, or $576.33 with the F2 reserved. If Copilot then competes with refreshes on that F2, you'll see it as throttling.

Copilot also answers only as well as the model it reads. Our Power BI semantic model guide covers what to document, and Power BI Copilot vs GetDot compares Copilot with an outside agent on the same model.

GetDot: AI on Power BI models and the warehouse without buying capacity for it

If the main reason you're pricing Fabric is to put an AI analyst in front of business users, there's another route. GetDot is an AI data analyst that connects to your Power BI semantic models and your warehouse, and answers questions in Teams, Slack, email or its web app, with the query it ran attached to every answer. Per our Power BI integration page:

• It reads each model's tables, measures, relationships and descriptions over the XMLA endpoint and runs DAX against your existing measures through the Power BI REST API's executeQueries endpoint, so answers use your measure definitions instead of re-deriving them.
• It can read which warehouse tables a model is built on and the exact number a report visual shows for a given period or slice.
• Access is read-only by default, and admins choose which Power BI actions Dot may take, per group.
• Since September 2026 it can also rebuild Power BI reports as Dot dashboards and compare them with the original tile by tile.

What it needs: the model has to sit in a workspace on Premium, Premium Per User or Fabric capacity, because Pro-only workspaces don't expose the XMLA endpoint, and the "Dataset Execute Queries REST API" tenant setting has to be on. So "without capacity" in practice means a PPU workspace, where the connecting account and everyone who opens that workspace needs PPU. If your models live in Pro workspaces, Dot can skip the model and query the warehouse directly: it connects to 45+ data sources, including Fabric warehouses and lakehouses through their SQL endpoint (where queries still run on that warehouse's capacity).

Dot is billed on credits, not seats, and every plan includes unlimited users (see pricing): Pro is $180 a month billed annually with 150 credits a month, and Team is $720 a month billed annually with 800 credits a month, plus SSO and row-level security. Dot queries your warehouse in place, its LLM providers run with zero data retention, and it is SOC 2 Type II and GDPR compliant.

One licensing point to check before you roll it out widely: Microsoft's multiplexing guidance says pooling access through another tool doesn't reduce the licenses you need. If Dot answers from a Power BI model for people who hold no Power BI license, confirm the terms with your Microsoft licensing contact. When Dot queries the warehouse directly, Power BI licensing doesn't come into it.

When Fabric and Copilot are the better choice

• You already run F64 or larger for report viewers, so Copilot only spends CUs you pay for anyway.
• Your data engineering lives in Fabric. Copilot works inside notebooks, pipelines and the warehouse editor; Dot doesn't.
• Your authors want AI inside Power BI Desktop and the report canvas, drafting report pages or writing DAX while they build. Dot works outside Power BI.
• You want one vendor and one contract, inside Microsoft's own compliance boundary.

Dot is the better fit when you're on Pro or PPU and don't want an F SKU just for AI, when questions arrive in Slack or Teams rather than inside reports, when answers need the Power BI model and the warehouse together, or when you'd rather pay for questions answered than for seats. For tools beyond these two, our Power BI alternatives guide compares ten of them. You can try Dot on the free plan (300 credits, no credit card) and connect a Power BI workspace in a few minutes.

Prices and requirements were checked on 29 October 2026 against the sources linked above. Fabric changes often; if anything here is out of date, tell us.

FAQ