Power BI Premium / Fabric Capacity Planning Guide 2026

Power BI Premium / Fabric Capacity Planning Guide 2026

Updated: April 25, 2026 · By: Errin O'Connor, Founder & Chief AI Architect, EPC Group · Reading time: 19 min

Power BI Premium and Microsoft Fabric F-SKU capacity sizing is more art than science. Get it wrong by 1 SKU tier and you waste $200K+/year. EPC Group has sized capacity for 80+ Fortune 500 deployments. This is the consolidated framework.

What capacity actually does

Power BI Premium / Fabric capacity = reserved compute that powers:

• Semantic model refresh
• Report query execution
• DirectQuery / DirectLake mode queries
• Dataflows + ETL pipelines
• AI workloads (Copilot for Fabric, Q&A)
• Real-Time Intelligence (Eventhouse / KQL)

SKU options (Q1 2026)

• F2 ($156/mo) — sandboxes, very small teams.
• F4 ($313/mo) — small teams.
• F8 ($625/mo) — single department.
• F16 ($1,250/mo) — multi-department mid-market.
• F32 ($2,500/mo) — small enterprise.
• F64 ($5,000/mo) — mid-enterprise; minimum for Power BI Premium per User free.
• F128 ($10,000/mo) — enterprise.
• F256 ($20,000/mo) — large enterprise.
• F512 ($40,000/mo) — Fortune 500.
• F1024 ($80,000/mo) — Fortune 100.
• F2048 ($160,000/mo) — global enterprise.

EPC Group's sizing methodology

Step 1: Workload telemetry (4 weeks)

Run Microsoft Fabric Capacity Metrics App for 4 weeks. Capture:

• Peak CU% utilization
• Peak background-vs-interactive split
• Refresh duration patterns
• DirectQuery / DirectLake query latency
• AI workload load

Step 2: Pattern analysis

• Steady-state: average CU% over a typical workday.
• Peak: highest 15-min window (often Monday 9am).
• Refresh windows: typically off-hours but can spike CPU during business hours.
• Sub-second SLA workloads: how much capacity needed for interactive sub-second queries.

Step 3: Right-sizing decision

EPC Group's rule: target peak CU% at 60-75%. Below 50% you are oversized; above 85% you risk throttling.

Step 4: Autoscale configuration

Fabric autoscale lets capacity expand temporarily for peak loads. Configure:

• Floor: 60% of expected steady-state.
• Ceiling: 150-200% of expected peak.
• Trigger: 80% sustained for 10+ minutes.

Step 5: Multi-region deployment

For global enterprises, deploy capacity in 2-3 regions matching user concentration. F-SKUs are region-specific. Use OneLake shortcuts to share data across regions.

Common mistakes

1. Sizing for peak only — 90% of capacity sits idle off-hours; autoscale solves this.
2. No telemetry baseline — sizing without 4 weeks of data is guessing.
3. Single SKU for entire enterprise — different workloads (refresh, interactive, AI) compete for same capacity.
4. Forgetting AI workloads — Copilot for Fabric + Q&A consume meaningful CU.
5. Ignoring DirectLake — DirectLake queries are typically 2-5× cheaper than Import; right-sizing differs.

Pricing comparison: Premium per User vs F-SKU

For 200 users:

• PPU: 200 × $20/mo = $4,000/mo
• F64: $5,000/mo (includes Power BI Premium per User free; unlimited users)

PPU breaks even at ~250 users vs F64. Above 250 users, F-SKU is cheaper. Below 100 users, PPU is simpler.

What capacity oversizing costs

Typical Fortune 500 oversize: F128 instead of F64. That's $5,000/mo × 12 = $60,000/year wasted. Across 5-year platform lifetime: $300K. EPC Group's right-sizing typically saves Fortune 500 clients $100-500K/year.

Frequently Asked Questions

How do we know if we are oversized?

Run Capacity Metrics App; check 30-day p95 CU%. If <50%, you are oversized.

How do we know if we are undersized?

Check throttling events; check refresh queue length; check DirectQuery query timeouts. Any sustained throttling = undersized.

Can we share capacity across business units?

Yes — single F-SKU can host multiple workspaces from different BUs. Use Capacity Metrics to allocate cost.

What about Power BI Pro for individual users?

Pro = $14/user/month for individual usage. Suitable for small teams without shared workspaces. Above 50 users, F-SKU is typically more cost-effective.

How does autoscale billing work?

Autoscale capacity is billed at 1.5× standard CU rate for the time spent expanded. Cheaper than oversizing baseline + idle.

Should we deploy multi-region capacity?

For global enterprises with >20% non-US users: yes. For US-only: single region (typically US East 2 or US Central) is fine.

What's the AI workload sizing rule?

Microsoft Copilot for Power BI / Fabric typically consumes 5-15% of base capacity at moderate use. Heavy use (AI Q&A across 1,000+ users) can consume 30-50%.

Does fabric capacity affect performance?

Yes — bigger capacity typically means lower query latency. EPC Group's pattern: F128 for sub-second interactive SLAs; F64 for 1-3 second SLAs; F32 for 5-10 second SLAs.

How often should we re-size?

Quarterly. Workloads grow; what was right at start of year is wrong by Q4 typically.

What's the maximum capacity?

F2048 is largest standard SKU. Above that requires Microsoft Enterprise Agreement custom commitment.

---

Sizing Power BI Premium / Fabric capacity? EPC Group's 4-week telemetry analysis + sizing methodology has saved clients $100-500K/year. Schedule a capacity sizing assessment.