Direct answer: at Azure Analysis Services public list prices on 10 November 2026 (eastus, USD, excluding tax and discounts), per instance-hour: Developer $0.132 ($96.36 for 730h), B1 $0.43 ($313.90), B2 $0.86 ($627.80), S0 $0.81 ($591.30), S1 $2.03 ($1,481.90), S2 $4.06 ($2,963.80), S4 $8.11 ($5,920.30), S8 v2 $10.38 ($7,577.40), S9 v2 $20.76 ($15,154.80). Scale-out replicas cost the same as the primary: S1 + 2 replicas = 3 × $1,481.90 = $4,445.70/month; S4 + 1 replica = $11,840.60. Pausing stops compute charges; S1 at 8h/day × 22 days (176h) is $357.28/month, S2 is $714.56. Per QPU-hour: S0–S4 about $0.0203, S8/S9 v2 about $0.0162; large models favor S8/S9. You cannot downgrade Standard to Basic or Basic to Developer—only upgrade or change within the same tier.
This guide is for data platform, BI developers, SRE, finance and procurement teams running or renewing Azure Analysis Services, evaluating Power BI/Fabric migration, or budgeting BI capacity. It separates official prices, editorial calculations and exclusions, covering Developer/Basic/Standard, QPU/memory, scale-out, pause, memory-based sizing, AAS vs Fabric/Power BI, hidden costs and a 90-day checklist. Prices come from the AAS pricing page, Azure Retail Prices API and Microsoft Learn (10 November 2026). This is not procurement advice.
(1) Billing is per instance/QPU-hour, and pausing stops compute; underlying Blob storage bills separately. (2) Every scale-out replica costs the same as the primary: S1 + 2 replicas is 3×; confirm query concurrency really needs it. (3) S8/S9 v2 have lower $/QPU and $/GB; large models favor S8/S9, small models stay flexible on S0–S4.
SKUs: QPU, memory and price
Developer (3 GB, no SLA; QPU specs differ between two docs—see table note) is for dev/test; Basic B1 (40/10) and B2 (80/16) suit small models, low concurrency and simple refresh but lack perspectives, multiple partitions and DirectQuery; Standard S0 (40/10), S1 (100/25), S2 (200/50), S4 (400/100), S8 v2 (640/200) and S9 v2 (1,280/400) fit production and larger models. Standard supports multiple partitions, DirectQuery and perspectives; S8/S9 are v2 with lower $/QPU/GB. Moves within a tier go both ways; cross-tier only upward—Standard→Basic and Basic→Developer are explicitly not allowed, while Basic→Standard and Developer→Basic are. Scale-out is Standard-only, with up to seven more query replicas.
| Instance | QPU | Memory | Hourly | 730h/month |
|---|---|---|---|---|
| Developer | 10 | 3 GB | $0.132 | $96.36 |
| B1 | 40 | 10 GB | $0.43 | $313.90 |
| B2 | 80 | 16 GB | $0.86 | $627.80 |
| S0 | 40 | 10 GB | $0.81 | $591.30 |
| S1 | 100 | 25 GB | $2.03 | $1,481.90 |
| S2 | 200 | 50 GB | $4.06 | $2,963.80 |
| S4 | 400 | 100 GB | $8.11 | $5,920.30 |
| S8 v2 | 640 | 200 GB | $10.38 | $7,577.40 |
| S9 v2 | 1,280 | 400 GB | $20.76 | $15,154.80 |
Sources: AAS pricing page and Retail API (eastus, 10 November 2026). Monthly = hourly × 730; excludes scale-out, storage, egress, tax and discounts. Developer has no SLA; its QPU is listed as 10 on the pricing page and D1 20 in the Learn overview—the table follows the pricing page, which does not affect the hourly price.
Per QPU / per GB: S8/S9 are better value
At 730 hours: S0 $/QPU is about $0.02025, S1 $0.0203, S2 $0.0203, S4 $0.020275; S8/S9 v2 are about $0.01622, roughly 20% lower than S0–S4. Per memory: S0 $0.081/GB-hour, S1 $0.0812, S2 $0.0812, S4 $0.0811; S8/S9 about $0.0519. So when the model exceeds 100 GB/400 QPU, S8 is often better value than S4; don't compare monthly fees alone—compare QPU/GB and actual memory needs.
| Instance | $/QPU-h | $/GB-h | Note |
|---|---|---|---|
| S0 | $0.020250 | $0.0810 | 40 QPU/10 GB |
| S1 | $0.020300 | $0.0812 | 100/25 |
| S2 | $0.020300 | $0.0812 | 200/50 |
| S4 | $0.020275 | $0.0811 | 400/100 |
| S8 v2 | $0.016219 | $0.0519 | 640/200, v2 savings |
| S9 v2 | $0.016219 | $0.0519 | 1,280/400 |
Editorial calculation: hourly ÷ QPU, ÷ GB; excludes scale-out, storage, tax and discounts. v2 feature/regional availability follows the official docs.
Pause and scale-out: save compute or add concurrency
Pausing stops compute charges; S1 at 8h/day × 22 days (176h) is $357.28, S2 $714.56, S4 $1,427.36. Underlying Blob storage bills via Azure Storage. Scale-out is Standard-only, adds up to seven query replicas, and each replica costs the same as the primary: S1 + 2 replicas = $4,445.70/month; S2 + 1 = $5,927.60; S4 + 1 = $11,840.60. Replicas add query concurrency and isolate refresh—not free read scaling; confirm concurrency and QPU utilization first.
| Scenario | Monthly | Note |
|---|---|---|
| S1 full month | $1,481.90 | 730h |
| S1 8h/day × 22d | $357.28 | 176h, paused otherwise |
| S2 8h/day × 22d | $714.56 | 176h |
| S1 + 2 replicas | $4,445.70 | 3× $1,481.90 |
| S2 + 1 replica | $5,927.60 | 2× $2,963.80 |
| S4 + 1 replica | $11,840.60 | 2× $5,920.30 |
Editorial calculation: hourly × hours; scale-out replicas same price; pause stops compute but Blob storage/egress continue; excludes tax and discounts.
Sizing by memory: 10/25/50/100/200/400 GB
Pick the smallest tier that fits model memory: 10 GB uses S0 ($591.30) or B1 ($313.90, but limited features); 25 GB S1 ($1,481.90); 50 GB S2 ($2,963.80); 100 GB S4 ($5,920.30); 200 GB S8 ($7,577.40); 400 GB S9 ($15,154.80). If the model is near the limit and refresh needs extra memory, leave 20%–30% headroom; otherwise refresh/query can fail. Developer (3 GB, no SLA) is not a production option.
| Model memory | Minimum Standard | Monthly | Note |
|---|---|---|---|
| 10 GB | S0 | $591.30 | B1 $313.90 but feature-limited |
| 25 GB | S1 | $1,481.90 | Common production start |
| 50 GB | S2 | $2,963.80 | Multi-partition/concurrency |
| 100 GB | S4 | $5,920.30 | Larger models |
| 200 GB | S8 v2 | $7,577.40 | Lower $/GB than S4 |
| 400 GB | S9 v2 | $15,154.80 | Largest standard tier |
Editorial summary: official QPU/memory tiers and 730h monthly; size by model, refresh peak and concurrency headroom; excludes tax and discounts.
AAS vs Fabric/Power BI: price the increment
AAS is a dedicated semantic-model service billed per instance/QPU, with pause, scale-out, DirectQuery and multiple partitions; Fabric/Power BI is a shared capacity/user platform. Before migrating, price Fabric capacity (F SKU, CU-hour) or Power BI PPU/Pro licences + storage/gateway/rework against AAS instance + replicas + operations. AAS offers mature tabular models, granular QPU and pause; Fabric offers shared capacity, OneLake and Power BI integration. Don't swap "AAS monthly" for "Fabric F64 monthly" directly—the capabilities and licences differ.
| Option | Billing | Fit | Note |
|---|---|---|---|
| AAS Standard | Instance/QPU-hour | Dedicated models, pause/scale-out | Replicas same price, Blob extra |
| AAS Basic | Instance-hour | Small models/low concurrency | No multi-partition/DirectQuery |
| Fabric F SKU | CU-hour | Shared capacity/OneLake/Power BI | Capacity + licence boundaries |
| Power BI PPU/Pro | Per user | Reports/self-service | Large model/refresh limits |
Editorial summary: AAS/Fabric/Power BI capabilities and licence boundaries differ; validate with the official calculator and a pilot, not monthly fees alone.
Selection matrix and hidden costs
Choose by scenario: dev/test uses Developer (3 GB, no SLA); small models/low concurrency use B1/B2 (feature-limited); production semantic models use S0–S4; large models/high QPU use S8/S9 v2; add scale-out only when query concurrency is short, but each replica costs the same; pause outside working hours. Seven hidden costs: replicas at full price, Blob storage/egress, storage while paused, no downgrades, Basic feature limits, migration rework/dual running, and Power BI licences. 90 days: weeks 1–2 inventory model size, QPU/concurrency and refresh windows; weeks 2–4 size S0–S9 by memory; weeks 4–6 evaluate pause/scale-out; weeks 6–8 compare Fabric/PPU increments; weeks 8–10 pilot migration/performance; weeks 10–12 reconcile instances, replicas and storage against the bill.
| Scenario | First choice | Check first |
|---|---|---|
| Dev/test | Developer | 3 GB, no SLA |
| Small model/low concurrency | B1/B2 | No multi-partition/DirectQuery |
| Standard production | S0–S4 | Memory headroom, refresh peak |
| Large model/high QPU | S8/S9 v2 | $/QPU, regional availability |
| Query concurrency short | Scale-out | Replica price, utilization |
| Off-hours | Pause | Blob storage continues |
Editorial summary, not a vendor recommendation; features/regions/downgrade limits follow the official docs.
FAQ
How is Azure Analysis Services billed?
Per instance-hour: Developer $0.132, B1 $0.43, B2 $0.86, S0 $0.81, S1 $2.03, S2 $4.06, S4 $8.11, S8 $10.38, S9 $20.76; 730h monthly is $96.36–$15,154.80. Scale-out replicas cost the same; pausing stops compute.
Am I charged while paused?
Pausing stops compute charges; underlying Blob storage bills via Azure Storage, and egress/other resources continue. S1 at 8h/day × 22 days is $357.28/month—good for off-hours.
How is scale-out billed?
Scale-out is Standard-only, up to seven query replicas, each at the same price as the primary. S1 + 2 replicas = 3× $1,481.90 = $4,445.70; S2 + 1 = $5,927.60; S4 + 1 = $11,840.60. Replicas add query concurrency/refresh isolation, not free read scaling.
S4 or S8 for a large model?
100 GB uses S4 ($5,920.30); 200 GB uses S8 ($7,577.40). S8/S9 have lower $/QPU/GB (about $0.0162/QPU-h, $0.0519/GB-h), so larger models get better value.
What is the difference between Basic and Standard?
Basic lacks perspectives, multiple partitions and DirectQuery; Standard supports them. Small models/low concurrency can use B1/B2, but production semantic models usually need S0 or above.
Can I downgrade?
Within a tier you can move both ways (S2↔S4); upgrades can cross tiers, but Standard→Basic and Basic→Developer are not allowed. Plan headroom to avoid a rebuild.
AAS or Fabric?
AAS bills per instance/QPU with pause/scale-out/multi-partition, suited to dedicated semantic models; Fabric bills CU capacity with shared OneLake/Power BI. Price Fabric capacity + licences + rework vs AAS instance + replicas + operations.
How do I reduce cost?
Five steps: (1) choose the smallest memory tier with 20%–30% headroom; (2) pause outside working hours; (3) add replicas only for real concurrency; (4) use S8/S9 v2 for large models; (5) evaluate migration/licence increments rather than switching platforms blindly.
Research method and limitations
Official facts (pricing page, Retail API, Learn): Developer/Basic/Standard QPU/memory and hourly prices; scale-out replicas same price; pause stops compute; no downgrades; AAS uses Blob storage/Entra ID. Editorial calculations: 730h, 176h = 8h × 22d, $/QPU, $/GB, replica multiples.
Seven limitations: (1) only eastus public list prices, excluding tax/contract discounts; (2) model size/concurrency/refresh are editorial assumptions; (3) excludes Blob storage/egress, migration rework, Power BI licences and tax; (4) no real refresh/query/concurrency benchmark; (5) v2 regional availability is not checked one by one; (6) Fabric/Power BI are framed as meter reminders, not same-workload quotes; (7) prices/features change—recheck official pages and bills before purchase.
Sources and basis
- Microsoft Azure, Azure Analysis Services Pricing: azure.microsoft.com/en-us/pricing/details/analysis-services. Developer/Basic/Standard, QPU/memory, scale-out, pause, downgrade limits. Accessed 10 November 2026.
- Azure Retail Prices API (productName contains Analysis Services, eastus): prices.azure.com/api/retail/prices. $0.132/$0.43/$0.86/$0.81/$2.03/$4.06/$8.11/$10.38/$20.76. Accessed 10 November 2026.
- Microsoft Learn, Azure Analysis Services overview: analysis-services-overview. Blob storage, Entra ID, semantic models. Accessed 10 November 2026.
- Microsoft Learn, Azure Analysis Services scale-out: analysis-services-scale-out. Scale-out replicas, query concurrency/refresh isolation. Accessed 10 November 2026.
Data cut-off 10 November 2026; prices/availability follow Microsoft official pages and bills; this is not procurement, commercial or legal advice.