Direct answer: at Azure Managed Instance for Apache Cassandra public list prices on 11 November 2026 (eastus, USD, excluding tax and discounts), cost is not per request or RU—it is nodes × VM SKU hourly price + P30 disks per node + backup. Using the common production assumption of 3 nodes and 4 P30 disks per node: D8s v5 (8 vCPU/32 GB) is $0.48/node/hour = $350.40/month/node, so 3 nodes cost $1,051.20; P30 at $135.17/disk/month × 12 = $1,622.04; disk mounts at $6.57/disk/month × 12 = $78.84; total $2,752.08/month. D16s v5 is $3,803.28/month, E8s v5 $3,080.58; 6 D8s v5 nodes $5,504.16; 6 P30 per node $3,602.52. Backup is $0.10/GB/month extra (1 TB ≈ $100/month). Microsoft says there is no licence fee; nodes/disks are selectable, and 4 P30 disks are a "strongly recommended" minimum.
This guide is for architects, DBAs, SRE, finance and procurement migrating self-hosted Cassandra to managed service, comparing Managed Instance for Apache Cassandra with Cosmos DB Cassandra/self-hosted, or budgeting nodes/disks for an existing cluster. It separates official prices, editorial calculations and exclusions, covering billing components, VM SKU prices, 3-node/multi-node TCO, P30 and mount fees, backup, replication-factor/50%-disk assumptions, hidden costs and a 90-day checklist. Prices come from the Managed Instance for Apache Cassandra pricing page, Azure Retail Prices API and Microsoft Learn introduction/FAQ/create docs (11 November 2026). This is not procurement advice.
(1) The bill is "nodes × VM hours + P30 per node + backup", not requests; 3 nodes/4 P30 with D8s v5 is about $2,752/month. (2) Disks often cost more than compute: 4 P30/node = $566.96/node/month, $1,700.88 for 3 nodes, close to the $1,051.20 for three D8s v5 nodes. (3) Four P30 disks are the "strongly recommended" minimum, and Microsoft recommends only 50% sustained disk use; size raw disks at 2× data, then add RF=3.
Billing components: nodes, disks, mounts, backup
The official model is on-demand, instance-based, with no licence fee: you choose node count, VM SKU/cores/memory and P30 disks per node. Four components: compute (VM SKU hourly × nodes), P30 data disks ($135.17/disk/month), disk mounts ($6.57/disk/month) and backup ($0.10/GB/month, full backups to Azure Storage). There is no separate request/RU fee; nodes and disks bill every hour they exist, and stopping the workload does not automatically pause compute.
| Component | Unit price | Billing unit | Note |
|---|---|---|---|
| Compute (VM SKU) | Per SKU hourly | Node/hour | D8s v5 0.48, D16s v5 0.96, E8s v5 0.63 |
| P30 data disk | $135.17 | Disk/month | At least 4 per node recommended |
| Disk mount | $6.57 | Disk/month | Retail API meter |
| Backup storage | $0.10 | GB/month | Full backup to Azure Storage |
| Licence fee | $0 | — | Microsoft states no licence fee |
Sources: pricing page, Retail API and Learn introduction/create docs (11 November 2026). Compute monthly estimates use SKU hourly × 730; P30 and mounts per disk; excludes tax and discounts.
VM SKUs: monthly cost per node
At 730 hours/month: D8s v5 (8 vCPU/32 GB) $0.48/h = $350.40/month/node; D16s v5 (16/64) $0.96/h = $700.80; E8s v5 (8/64) $0.63/h = $459.90; L8s v3 (8/64, local disks) $0.87/h = $635.10. D-series is general purpose, E-series has more memory, L-series has local disks and suits read-intensive/write-through caching (Microsoft marks write-through caching as preview, not for production). Size by memory/CPU and disk throughput, not vCPU price alone; Cassandra usually cares more about memory and disk IO.
| VM SKU | vCPU/RAM | Hourly | Monthly/node | 3-node compute |
|---|---|---|---|---|
| D8s v5 | 8 / 32 GB | $0.48 | $350.40 | $1,051.20 |
| D16s v5 | 16 / 64 GB | $0.96 | $700.80 | $2,102.40 |
| E8s v5 | 8 / 64 GB | $0.63 | $459.90 | $1,379.70 |
| L8s v3 | 8 / 64 GB | $0.87 | $635.10 | $1,905.30 |
| D8s v5 ×6 nodes | 8 / 32 GB | $0.48 | $350.40 | $2,102.40 (6 nodes) |
Editorial calculation: hourly × 730; 3 nodes = per node × 3; excludes P30, mounts, backup, tax and discounts. The pricing page lists hourly prices per VM SKU; verify the bill.
3-node TCO: compute + 4 P30/node
With 3 nodes (common RF=3 budget assumption) and 4 P30 per node: disks $135.17×4 = $540.68/node/month, mounts $6.57×4 = $26.28/node/month, disk total $566.96/node/month, 3 nodes $1,700.88. Add compute: D8s v5 $2,752.08/month, E8s v5 $3,080.58, D16s v5 $3,803.28. Add 1 TB backup at about $100/month, so the D8s v5 case is about $2,852/month. Disk costs are close to or above compute—the most underestimated part of a Cassandra budget.
| 3-node option | Compute/month | 4 P30+mounts/month | Total/month | +1 TB backup |
|---|---|---|---|---|
| D8s v5 | $1,051.20 | $1,700.88 | $2,752.08 | $2,852.08 |
| E8s v5 | $1,379.70 | $1,700.88 | $3,080.58 | $3,180.58 |
| D16s v5 | $2,102.40 | $1,700.88 | $3,803.28 | $3,903.28 |
| L8s v3 | $1,905.30 | $1,700.88 | $3,606.18 | $3,706.18 |
Editorial calculation: 4×(135.17+6.57)=566.96/node; 3 nodes=1,700.88; backup at 0.10/GB×1,024 GB≈102.40 (table uses 100); excludes tax and discounts.
Scaling: more nodes vs more disks
Two scaling paths: add nodes for throughput/resilience, add P30 for per-node capacity. For D8s v5: 6 nodes + 4 P30 = compute $2,102.40 + disks $3,401.76 = $5,504.16/month; 3 nodes + 6 P30 = $1,051.20 + 3×($135.17+$6.57)×6 = $3,602.52/month; 3 nodes + 8 P30 = $4,452.96/month. If the bottleneck is capacity, disks are usually cheaper; if CPU/throughput/replicas, nodes help. Microsoft recommends only 50% disk use, so plan raw capacity×50%, then apply RF=3.
| Option | Compute/month | Disks+mounts/month | Total/month |
|---|---|---|---|
| 3 nodes × 4 P30 | $1,051.20 | $1,700.88 | $2,752.08 |
| 6 nodes × 4 P30 | $2,102.40 | $3,401.76 | $5,504.16 |
| 3 nodes × 6 P30 | $1,051.20 | $2,551.32 | $3,602.52 |
| 3 nodes × 8 P30 | $1,051.20 | $3,401.76 | $4,452.96 |
| 6 nodes × 6 P30 | $2,102.40 | $5,102.64 | $7,205.04 |
Editorial calculation: per-node disks = disks × (135.17+6.57); nodes × per-node; excludes backup, tax and discounts.
Replication factor, 50% disk and backup
The official create doc states: at least 4 P30 per node is "strongly recommended"; plan for only 50% sustained disk utilization to leave room for tombstones and system use; backups temporarily consume local disk before transfer to Blob. Cassandra commonly uses RF=3; 3 nodes is a common budget starting point, not proof that 3 nodes survive every failure—cross-AZ/multi-datacenter needs more nodes and networking. Backup is $0.10/GB/month, full backups to Azure Storage; frequency and retention drive cost. The service supports Cassandra versions up to 5.0; check version/extension compatibility before migrating.
| Item | Official basis | Budget impact |
|---|---|---|
| P30 per node | At least 4 (strongly recommended) | 4×141.74=$566.96/node/month |
| Disk utilization | Plan for 50% sustained use | Raw capacity roughly 2× data |
| Replication factor | Common RF=3 (editorial) | Logical data × RF raw copies |
| Backup | Full to Azure Storage, 0.10/GB/month | 1 TB≈$100/month; retention adds |
| Version | Supports up to Cassandra 5.0 | Check compatibility before migration |
| AZ/multi-DC | Optional, region-dependent | Nodes/network extra |
Sources: Learn create/introduction docs and Retail API (11 November 2026). RF=3 is an editorial budget assumption; actual replication follows the business; backup priced per GB-month.
Selection: managed Cassandra vs self-hosted vs Cosmos DB
Managed Cassandra fits teams with an existing Cassandra app/driver, needing managed repair/health/scaling and CQL/multi-datacenter semantics without running JVM/disk/repair; cost is nodes+disks, growing with scale. Self-hosted Cassandra fits teams with strong operations, reservations/spot/local disks and very large clusters; you own VMs, disks, backup, repair, monitoring and labour. Cosmos DB Cassandra API fits new apps wanting RU/serverless elasticity and accepting API/consistency differences; don't treat a CQL workload as equivalent—validate compatibility and cost first (see the Cosmos DB cost guide).
| Scenario | First choice | Check first |
|---|---|---|
| Existing Cassandra/CQL | Managed Cassandra | Version ≤5.0, driver compatibility |
| Managed repair/monitoring needed | Managed Cassandra | Node/disk TCO |
| Strong ops team, large scale | Self-hosted | VMs/disks/labour/repair |
| New app, elasticity first | Cosmos DB Cassandra API | RU/consistency/compatibility |
| Capacity bottleneck | Add P30 | 50% utilization, RF |
| Throughput/replica bottleneck | Add nodes/up-SKU | CPU/memory/disk IO |
| Cross-AZ/multi-region | Multi-datacenter | Node/network cost |
Editorial summary, not a vendor recommendation; compatibility/prices follow official docs and bills; Cosmos DB cost is covered in its own article.
Hidden costs and 90-day checklist
Eight hidden costs: P30 per disk, mount fees, backup storage/retention, 50%-disk planning doubling raw capacity, RF=3 copies, cross-AZ/multi-DC nodes, no automatic compute pause when stopped, and migration/dual-running/labour. 90 days: weeks 1–2 inventory data size, RF, disks per node, peak IO/CPU; weeks 2–3 price 3/6 nodes and 4/6/8 P30; weeks 3–4 check version ≤5.0 and driver/extension compatibility; weeks 4–6 pilot migration/dual run; weeks 6–8 load-test repair/scaling/failover; weeks 8–10 optimize disk utilization and backup retention; weeks 10–12 reconcile the bill and set node/disk/backup alerts.
| Cost item | Trigger | Control |
|---|---|---|
| Compute | Nodes × SKU × hours | Size SKU by peak/replicas |
| P30 | Disks per node | Size at 50% utilization |
| Mounts | Per disk/month | Include with disk count |
| Backup | $0.10/GB/month | Retention/frequency |
| RF/copies | RF=3 etc. | Business resilience |
| Cross-AZ/multi-region | Multi-datacenter | Deploy on demand |
| Stopped workload | No automatic pause | Delete/scale down |
| Migration/labour | Dual run/rework | Rehearse early |
Editorial summary: resources outside the price list bill as other Azure services; networking/monitoring/labour are not included.
FAQ
How is Managed Instance for Apache Cassandra billed?
Nodes × VM SKU hourly + P30 disks per node ($135.17/disk/month) + disk mounts ($6.57/disk/month) + backup ($0.10/GB/month); no licence fee, no per-request/RU charge.
What does a 3-node cluster cost per month?
With D8s v5 and 4 P30/node: compute $1,051.20 + disks/mounts $1,700.88 = $2,752.08/month; with 1 TB backup about $2,852. D16s v5 is about $3,803.28; E8s v5 about $3,080.58.
Why do disks cost more than compute?
4 P30/node = 4×($135.17+$6.57)=$566.96/node/month, $1,700.88 for 3 nodes; three D8s v5 nodes are $1,051.20. Cassandra's capacity/replication needs make disks the main cost.
Are 4 P30 disks mandatory?
The create doc calls at least 4 "strongly recommended", not a hard minimum; fewer may hurt throughput/capacity/repair. This article uses 4 as the baseline; adjust to data/IO.
What does 50% disk utilization mean?
Microsoft recommends planning for only 50% sustained disk use, leaving room for tombstones, system use and temporary backup files. Raw disk capacity is roughly 2× expected data, then apply RF=3 copies.
How much is backup?
Backup storage is $0.10/GB/month, full backups to Azure Storage; 1 TB is about $100/month. More frequent/longer retention costs more; backups temporarily consume local disk before transfer.
Managed Cassandra or Cosmos DB Cassandra API?
For existing Cassandra/CQL and managed repair/multi-datacenter, choose managed Cassandra; for a new app wanting elasticity/RU billing and accepting API/consistency differences, evaluate Cosmos DB Cassandra API. Validate compatibility and same-workload pricing, not unit price alone.
How do I reduce cost?
Five steps: (1) size disks precisely at 50% utilization and RF, avoid overbuying; (2) add P30 for capacity, nodes for throughput; (3) choose the right D/E/L SKU rather than overprovisioning; (4) optimize backup frequency/retention; (5) compare self-hosted/reserved for stable long-term clusters, including labour.
Research method and limitations
Official facts (pricing page, Retail API, Learn introduction/FAQ/create docs): node/VM SKU billing, no licence fee, P30 $135.17/disk/month, mount $6.57/disk/month, backup $0.10/GB/month, 4 P30 strongly recommended, 50% disk utilization, Cassandra 5.0 support, managed repair/monitoring/scaling. Editorial calculations: 3 nodes/RF=3, 4 P30/node, 730 hours/month; D8s/D16s/E8s/L8s monthly; 6-node/6–8 P30 scaling; 1 TB backup≈$100/month.
Seven limitations: (1) eastus USD list prices only, excluding tax/contract discounts/regional differences; (2) node count, SKU, disk count, RF and backup volume are editorial assumptions; (3) excludes network/monitoring/labour, cross-AZ/multi-region traffic and migration dual running; (4) no real repair/scaling/failover benchmark; (5) the pricing page lists per-VM-SKU hourly prices while Retail API meter names contain vCore—verify the bill; (6) L-series write-through caching is preview and not for production; (7) prices/version support change—recheck official pages and bills before purchase.
Sources and basis
- Microsoft Azure, Azure Managed Instance for Apache Cassandra Pricing: azure.microsoft.com/en-us/pricing/details/managed-instance-apache-cassandra. VM SKU hourly prices, P30/mounts, backup, no licence fee. Accessed 11 November 2026.
- Azure Retail Prices API (productName contains Cassandra, eastus): prices.azure.com/api/retail/prices. D8s v5 0.48, D16s v5 0.96, E8s v5 0.63, L8s v3 0.87, P30 135.17, Disk Mount 6.57, Backup 0.10/GB. Accessed 11 November 2026.
- Microsoft Learn, Introduction to Azure Managed Instance for Apache Cassandra: introduction. Managed repair/monitoring/scaling, no licence fee, Cassandra 5.0. Accessed 11 November 2026.
- Microsoft Learn, Create a cluster from the portal: create-cluster-portal. 4 P30 strongly recommended, 50% disk utilization, backup local disk use, AZ/node/SKU selection. Accessed 11 November 2026.
Data cut-off 11 November 2026; prices/features follow Microsoft official pages and bills; this is not procurement, architecture or legal advice.