Microsoft Fabric F0 Pricing Calculator: On-Demand Billing, Project Osmos & F2–F64 Break-Even
Compare Microsoft Fabric’s new zero-provisioned F0 SKU (on-demand CU-hour billing) against Paused Pay-As-You-Go and 1-Year Reserved F2–F8192 capacities. Model Project Osmos data engineering agent token-to-CU burn (GPT-5.5 / GPT-5.4) and test how 3× Capacity Overage and rolling 24-hour CU spend limits impact your Azure bill.
Built by A.J. · Last verified: · Checked against Microsoft Learn Project Osmos Consumption Rates, Fabric Capacity Overage (3× PAYG), and FabCon Europe 2026 F0 On-Demand Specifications.
🧮 Interactive Fabric F0, Project Osmos & Overage FinOps Simulator
Step 1: Workload Consumption & F0 Multiplier
Un-smoothed CU-hours your jobs actually burn per month.
Microsoft prices F0 on-demand as a multiplier of PAYG ($0.18/CU-hr US).
730 hrs = Always-On (24/7). 220 hrs ≈ 10 hrs/day × 22 workdays.
Step 2: OneLake Storage & Power BI Licensing
Billed separately (~$26/TB US, €20/TB EU, £18/TB UK).
On F0 and below F64, viewers need Pro (~$14/user/mo).
F0 vs. Provisioned F-SKU Decision Verdict
1. F0 Zero-Provisioned (On-Demand)
$287/mo
$95 compute + $52 storage + $140 PBI
2. F2 (1-Year Reserved, -41%)
$348/mo
$156 compute + $52 storage + $140 PBI
3. F2 Paused PAYG (220 hrs/mo)
$271/mo
$79 compute + $52 storage + $140 PBI
4. F2 Always-On PAYG (730 hrs)
$455/mo
$263 compute + $52 storage + $140 PBI
Selected SKU Monthly CU Pool (730 hrs)1,460 CU-hours
Your Implied Capacity Utilization28.8% of F2
F0 Break-Even vs. 1-Year Reserved SKU693 CU-hrs/mo (47.5%)
F0 Break-Even vs. Always-On PAYG SKU1,168 CU-hrs/mo (80.0%)
Recommended 24-Hour Spend Limit (1.5× avg day)21 CU-hours / 24h
Step 1: Project Osmos Billing Profile & Context Window
Official Microsoft Learn conversion rates from
data-engineering-agent-consumption.
>272,000 tokens triggers the Long-Context tier!
Note: Pro profiles do not price cached input separately.
E.g. 250 steps/mo (normal) or 2,500 steps (runaway loop).
Project Osmos CU & Dollar Cost Breakdown
Project Osmos Compute Cost
$359
At $0.18/CU-hr across 250 agent steps
Total CU-Hours Consumed
1,996 CU-hrs
7,185,763 CU-seconds total
Active Context Tier
Short (≤272K)
180,000 prompt tokens / step
If Switched to GPT-5.4-mini
$54
Saves $305 (85% less CU burn)
Uncached Input Rate (CU-s / 1K tokens)139.83 CU-s
Cached Input Rate (CU-s / 1K tokens)13.98 CU-s
Output Token Rate (CU-s / 1K tokens)838.98 CU-s
Share of an Always-On F2 Capacity (1,460 CU-hrs)136.7% of F2
Cost if Billed on 3× Capacity Overage Meter$1,078
Step 1: Provisioned Capacity & 24-Hour Overage Limit
Microsoft’s default on new capacities is 25% per 24h.
How many days/month your workloads spill into 3× overage.
Because Fabric checks thresholds periodically and lets running jobs finish, in-flight jobs can exceed your 24h cap!
3× Capacity Overage vs. SKU Upsize Verdict
Monthly 3× Overage Bill Shock
+$486/mo
900 overage CU-hrs at $0.54/CU-hr (3× PAYG)
Base SKU + Overage Total
$1,111/mo
$625 base F8 Reserved + $486 overage
Next SKU Up (1-Yr Reserved)
$1,251/mo
F16 Reserved gives +5,840 extra CU-hrs/mo
If Offloaded to Spark On-Demand (1×)
$787/mo
$625 base + $162 Spark on-demand (saves $324)
3× Capacity Overage Rate (Separate Azure Meter)$0.54 / CU-hour
Configured 24h Threshold Allowance48 CU-hours / day
Un-Capped In-Flight Overshoot per Burst Day12 CU-hours / day
Break-Even Days Where Upsizing SKU (Reserved) Wins19.3 burst days / mo
⚠️ Critical FabCon Europe 2026 Governance Warning: Microsoft’s new Zero-Provisioned F0 SKU and Per-Category On-Demand Billing meter consumption in CU-hours without 24-hour smoothing, while Capacity Overage (GA) is enabled by default on new F-capacities at 3× the Pay-As-You-Go rate ($0.54/CU-hour in US regions). Crucially, both the 24-hour on-demand spend limit and the capacity overage threshold are soft spending thresholds, not hard kill-switches: when reached, new operations pause, but in-flight operations already running continue to completion and are billed in full.
Quick Answer: How does Microsoft Fabric F0 & On-Demand Billing work?
Microsoft Fabric F0 is a zero-provisioned entry SKU with $0 upfront compute charge where supported workloads (such as Spark, Data Warehouse, and Data Factory) bill per consumed Capacity Unit hour (CU-hour) on demand. Because a 1-Year Reserved F-SKU saves roughly 41% off the Pay-As-You-Go rate ($0.107/CU-hr vs. $0.18/CU-hr in US regions), F0 is cheapest when your average capacity utilization stays below ~40% to 59% (depending on the category multiplier), while steady production workloads and estates needing F64 free Power BI viewers remain cheaper on a 1-year reservation.
1. Microsoft Fabric F0 vs. F2–F8192 Capacity Pricing Table (2026–2027)
Until September 2026, every team adopting Microsoft Fabric after the 60-day trial had to provision at least an F2 capacity (2 CUs)—paying for compute either 24/7 on a reservation or manually pausing and resuming a Pay-As-You-Go SKU via Azure runbooks. With the announcement of Fabric F0 (zero-provisioned), per-category On-Demand Billing, and the new F4096 and F8192 mega-SKUs at FabCon Europe 2026, architects now have four distinct ways to pay for Fabric compute:
| Fabric SKU / Billing Model | Capacity Units (CUs) | Monthly CU-Hour Pool | US East PAYG (Always-On) | US East 1-Yr Reserved (~41% Off) | Best For / Power BI Viewer Rule |
|---|---|---|---|---|---|
| F0 (Zero-Provisioned) | 0 Upfront (On-Demand) | Pay per CU-hr used (No smoothing) | $0 upfront + On-Demand CU-hrs | N/A (Consumption billed) | PoCs, dev/test, spiky workloads under ~45%–59% utilization. Requires Power BI Pro ($14/mo). |
| F2 | 2 CUs | 1,460 CU-hrs/mo | ~$263 / mo ($0.36/hr) | ~$156 / mo | Smallest production lakehouse. Viewers need Power BI Pro ($14/mo). |
| F4 | 4 CUs | 2,920 CU-hrs/mo | ~$526 / mo ($0.72/hr) | ~$313 / mo | Small departmental ETL & warehouse. Viewers need Pro. |
| F8 | 8 CUs | 5,840 CU-hrs/mo | ~$1,051 / mo ($1.44/hr) | ~$625 / mo | Growing analytics team; consider offloading bursty Spark to On-Demand. |
| F16 | 16 CUs | 11,680 CU-hrs/mo | ~$2,102 / mo ($2.88/hr) | ~$1,251 / mo | Mid-market medallion lakehouse + scheduled refreshes. |
| F32 | 32 CUs | 23,360 CU-hrs/mo | ~$4,205 / mo ($5.76/hr) | ~$2,501 / mo | High-concurrency analytics with <200 Power BI report viewers. |
| F64 (Free Viewer Cliff) | 64 CUs | 46,720 CU-hrs/mo | ~$8,410 / mo ($11.52/hr) | ~$5,003 / mo | Unlimited Free Power BI Viewers (saves $14/mo per viewer; breaks even at ~179+ viewers vs F32). |
| F128 – F2048 | 128 – 2,048 CUs | 93,440 – 1,495,040 CU-hrs | $16,819 – $269,107 / mo | $10,005 – $160,085 / mo | Enterprise multi-department estates with free Power BI viewers. |
| F4096 & F8192 (New GA) | 4,096 – 8,192 CUs | 2,990,080 – 5,980,160 CU-hrs | ~$538,214 – $1,076,429 / mo | ~$320,170 – $640,340 / mo | Mega-scale global estates (Generally Available since Sept 2026). |
2. Project Osmos (Fabric Data Engineering Agent) Official Token-to-CU Rate Table
One of the most important cost drivers introduced in late 2026 is Project Osmos—Microsoft Fabric’s autonomous data engineering agent that writes, tests, and repairs lakehouse notebooks and pipelines. Under Microsoft Learn’s Project Osmos capacity consumption documentation, language model tokens are converted directly into Fabric Capacity Unit seconds (CU-seconds) per 1,000 tokens and billed against the Copilot and AI background operation meter.
Watch out for two critical pricing mechanics in the official table below:
- The 272,000-Token Long-Context Cliff: Any request whose total prompt length exceeds 272,000 tokens (even 272,001 tokens, determined before splitting cached vs. uncached input) automatically jumps to the Long-Context billing profile—which doubles input CU consumption and increases output CU consumption by 50%!
- Pro Model Multiplier (6× to 18× Higher Burn): Switching an autonomous agent loop from
GPT-5.5toGPT-5.5-pro/GPT-5.4-proincreases output consumption from 838.98 CU-seconds/1K tokens to 5,033.90 CU-seconds/1K tokens (short context) or 7,550.85 CU-seconds/1K tokens (long context)—and Pro profiles do not offer a discounted cached-input rate.
| Project Osmos Billing Profile | Uncached Input (CU-s / 1K Tokens) | Cached Input (CU-s / 1K Tokens) | Output Tokens (CU-s / 1K Tokens) | CU-Hours per 1M Output Tokens | US Cost per 1M Output Tokens ($0.18 PAYG) |
|---|---|---|---|---|---|
| GPT-5.4-nano | 5.5932 | 0.5593 | 34.9576 | 9.71 CU-hrs | $1.75 |
| GPT-5-mini | 6.9915 | Not separately priced | 55.9322 | 15.54 CU-hrs | $2.80 |
| GPT-5.4-mini | 20.9746 | 2.0975 | 125.8475 | 34.96 CU-hrs | $6.29 |
| GPT-5.1 | 34.9576 | Not separately priced | 279.6610 | 77.68 CU-hrs | $13.98 |
| Generic Profile | 100.0000 | Not separately priced | 400.0000 | 111.11 CU-hrs | $20.00 |
| GPT-5.4, short context (≤272K) | 69.9153 | 6.9915 | 419.4915 | 116.53 CU-hrs | $20.97 |
| GPT-5.4, long context (>272K) | 139.8305 | 13.9831 | 629.2373 | 174.79 CU-hrs | $31.46 |
| GPT-5.5, short context (≤272K) | 139.8305 | 13.9831 | 838.9831 | 233.05 CU-hrs | $41.95 |
| GPT-5.5, long context (>272K) | 279.6610 | 27.9661 | 1,258.4746 | 349.58 CU-hrs | $62.92 |
| GPT-5.4-pro / GPT-5.5-pro, short (≤272K) | 838.9831 | Not separately priced | 5,033.8983 | 1,398.31 CU-hrs | $251.70 ($755 on 3× Overage) |
| GPT-5.4-pro / GPT-5.5-pro, long (>272K) | 1,677.9661 | Not separately priced | 7,550.8475 | 2,097.46 CU-hrs | $377.54 ($1,133 on 3× Overage) |
3. How to Prevent Runaway Bills on F0, On-Demand & Capacity Overage
Whether you are adopting the new F0 zero-provisioned SKU or enabling per-category On-Demand Billing on an existing F2–F64 capacity, apply this 4-step FinOps checklist before running production or AI agent workloads:
- Audit the Default 25% Capacity Overage Setting Today: According to Microsoft Learn, when you create a new Fabric capacity, Capacity Overage is enabled by default with a 25% rolling 24-hour allowance billed at 3× the Pay-As-You-Go rate ($0.54/CU-hr in US). If you prefer temporary throttling over paying 3× rates, disable Capacity Overage in the Azure / Fabric Admin portal immediately.
- Set Rolling 24-Hour CU-Hour Spend Limits Below Your Alert Ceiling: Because in-flight operations continue to completion after a 24-hour spend limit is triggered, set your rolling 24-hour CU-hour limit to 60%–70% of your true maximum daily budget so an in-flight Spark, Warehouse, or Project Osmos job cannot push your invoice over budget.
- Offload Variable Spark Workloads to On-Demand Billing for Spark (1× PAYG): Instead of sizing your entire F-SKU for a 2-hour nightly Spark peak—or paying 3× Capacity Overage—enable On-Demand Billing for Apache Spark (formerly Autoscale Billing for Spark), which bills active Spark job compute at the standard 1× PAYG rate (0.5 CU-hour per vCore-hour) under a hard maximum CU quota.
- Use the Hybrid Baseline + On-Demand Architecture: Reserve an F-SKU (saving ~41%) sized strictly for your steady 24/7 baseline load, and move spiky billing categories to On-Demand Billing with a tight 24-hour CU-hour spend limit.
Frequently Asked Questions: Microsoft Fabric F0, On-Demand & Project Osmos
What is Microsoft Fabric F0 and how is it different from F2?
Microsoft Fabric F0 (“Fabric zero-provisioned”), announced at FabCon Europe on September 29, 2026, is a new entry SKU with $0 upfront compute-capacity charge. Unlike an F2 capacity—which bills a fixed hourly rate ($0.36/hr PAYG or ~$156/mo reserved) for the entire time the capacity is running—F0 uses on-demand billing by default, charging only for the exact CU-hours your workloads consume without 24-hour smoothing.
When should I choose F0 On-Demand vs. an F2 1-Year Reservation?
Because a 1-year reservation on F2 costs ~$156/month for 1,460 CU-hours (~$0.107/CU-hr), F0 On-Demand is cheaper whenever your monthly consumption is below 867 CU-hours (at a 1.0× PAYG rate of $0.18/CU-hr) or below 578 CU-hours (at a 1.5× on-demand multiplier). Once your steady workload exceeds roughly 40% to 59% of an F2 capacity—or if you can pause an F2 PAYG capacity overnight and on weekends—a provisioned F2 becomes cheaper.
Why did Project Osmos consume so many Capacity Units on my lakehouse?
Project Osmos converts LLM input and output tokens into Fabric CU-seconds. If an autonomous agent loop uses a Pro reasoning profile (such as GPT-5.5-pro or GPT-5.4-pro) or if the conversation history exceeds 272,000 prompt tokens (triggering the Long-Context tier), each 1,000 output tokens consumes up to 7,550.85 CU-seconds (2.1 CU-hours). Capping context windows below 272,000 tokens and routing routine tasks to GPT-5.4-mini cuts CU burn by over 85%.
Is Capacity Overage really billed at 3× the Pay-As-You-Go rate?
Yes. Microsoft Learn confirms that Capacity Overage—which prevents throttling when your capacity exceeds 100% utilization—is billed on a separate Azure meter at three times (3×) the standard Pay-As-You-Go CU-hour rate ($0.54/CU-hr in US regions) and is not eligible for reservation discounts. Check your capacity settings in the Admin portal, as overage is enabled by default on newly created capacities at a 25% daily threshold.
Does setting a 24-hour CU spend limit guarantee my bill won’t exceed that number?
No. Both the F0/On-Demand rolling 24-hour spend limit and the Capacity Overage threshold pause new operations once the threshold is crossed, but allow in-flight operations already running to finish and bill in full. Always set your 24-hour CU-hour threshold below your hard budget ceiling to leave room for in-flight job completion.
Official Primary Sources & FinOps References
- Microsoft Learn: Fabric Data Engineering Agent (Project Osmos) Capacity Consumption Rates (Published September 28, 2026)
- Microsoft Learn: Capacity Overage in Microsoft Fabric (3× PAYG Billing & Default 25% Threshold)
- Microsoft Learn: On-Demand Billing for Apache Spark in Microsoft Fabric (0.5 CU-Hour Rate)
- Solv Systems: Microsoft Fabric F0 and On-Demand Billing — What Changes and When (FabCon Europe 2026 Analysis)
- Daatlers: Microsoft Fabric Pricing Guide — F0 to F2048 Capacity Costs & Smoothing Rules (October 2026)
Explore More Cloud, Data & AI Cost Calculators
⚠️ FinOps & Pricing Disclaimer: This calculator provides directional cloud architecture and FinOps estimates based on Microsoft’s published Azure Fabric rate cards (East US, Sweden Central, and UK South), Microsoft Learn’s Project Osmos token consumption tables, and FabCon Europe 2026 announcements current as of October 2026. Because F0 and per-category On-Demand Billing are rolling out in preview and category multipliers vary by workload and Enterprise Agreement, always verify live regional rates in the Azure Pricing Calculator and configure rolling 24-hour spend limits in the Fabric Admin Portal before running production workloads. Ultimate Info Guide is an independent publication and is not affiliated with or endorsed by Microsoft Corporation.
Advertisements