AzureAzure Synapse AnalyticsAnalytics

`azurerm_synapse_workspace` cost estimation

Unified data warehouse + Spark + serverless SQL + Data Factory pipelines. Dedicated SQL pool from $1.20/DWU-hour. Spark pool $0.143/vCore-hour. Serverless SQL $5/TB scanned. Workspace itself is free.

Azure Synapse Analytics combines data warehousing, Spark, serverless SQL, and pipelines in one platform. The azurerm_synapse_workspace creates the workspace; SQL pools, Spark pools, and pipelines are configured as separate resources. The workspace itself is free; you pay for the underlying compute.

Dedicated SQL Pool (formerly SQL Data Warehouse): - DW100c: $1.20/hour ($876/month) - DW1000c: $12/hour ($8,760/month) - DW30000c: $360/hour ($263K/month) - DWU = Data Warehouse Unit. Higher DWU = more capacity.

Can be paused (no compute cost, only storage). Storage: $122.88/TB-month for the SQL pool.

Serverless SQL Pool: - $5 per TB scanned - No always-on cost - Right for ad-hoc analysis, exploration

Spark Pool: - $0.143/vCore-hour for memory optimized - $0.105/vCore-hour for compute optimized - Auto-pause when idle

Pipelines (Data Factory inside Synapse): - $1 per 1,000 activity runs (same as Data Factory) - DIU-based for Copy activities

A typical Synapse workload: - Dedicated SQL DW500c paused at night (10 hours active): $876 × 10/24 = $365/month - Serverless SQL exploring 500 GB/month: $2.50 - Spark pool 10 vCores running 2 hours/day: 10 × $0.143 × 60 = $86/month - Total: ~$453/month

Synapse vs Databricks: Synapse is broader (DW + Spark + SQL + pipelines in one platform). Databricks is Spark-focused with better dev experience. For pure Spark, Databricks usually wins. For data warehouse-led workloads, Synapse Dedicated SQL Pool is purpose-built.

c3x estimates Synapse based on pool configurations (sku for SQL, node_size for Spark).

Terraform example

A minimal but realistic configuration that C3X can estimate.

resource "azurerm_synapse_workspace" "main" {
  name                                 = "prod-synapse"
  resource_group_name                  = azurerm_resource_group.main.name
  location                             = azurerm_resource_group.main.location
  storage_data_lake_gen2_filesystem_id = azurerm_storage_data_lake_gen2_filesystem.main.id

  sql_administrator_login          = "sqladmin"
  sql_administrator_login_password = var.sql_password

  identity {
    type = "SystemAssigned"
  }

  tags = {
    Environment = "production"
  }
}

resource "azurerm_synapse_sql_pool" "main" {
  name                 = "main-dw"
  synapse_workspace_id = azurerm_synapse_workspace.main.id
  sku_name             = "DW500c"
  create_mode          = "Default"
  storage_account_type = "LRS"
}

Pricing dimensions

What you actually pay for when you provision azurerm_synapse_workspace.

Dimension	Unit	What's being charged
Dedicated SQL Pool (DWU)	per DWU-hour	Compute capacity for the data warehouse. DW100c is the smallest tier. $1.20/hour for DW100c
Serverless SQL	per TB scanned	Pay-per-query for ad-hoc analytics. No idle cost. $5 per TB scanned
Spark Pool vCores	per vCore-hour	Spark compute for big data processing. Auto-pauses when idle. $0.105-$0.143/vCore-hour
Dedicated SQL storage	per TB-month	Storage for Dedicated SQL Pool data. $122.88/TB-month
Pipelines	per 1,000 runs	Data Factory-equivalent pipeline orchestration inside Synapse. $1 per 1,000 activity runs

Optimization tips

Common ways to reduce azurerm_synapse_workspace cost without changing the workload.

Pause Dedicated SQL pools during idle hours

30-60% on Dedicated SQL

Pausing stops compute billing entirely. Only storage continues. A DW500c paused 14 hours/day saves ~58% of compute cost. Schedule pause/resume via Azure Automation or Logic Apps.

Use Serverless SQL for ad-hoc queries

100% on idle Dedicated SQL avoided

For occasional analysis, Serverless SQL bills only per query. Avoid spinning up Dedicated SQL for one-off explorations. Pay only for what you scan.

Right-size DWUs

$876+/month per DW100c removed

DWUs are linear cost. Many teams provision DW1000c+ 'for headroom' but actually use 10-20%. Monitor DWU usage; right-size to actual peak workload. Scaling DWUs up is online and fast.

Reserved Capacity for stable production

37-65% on commitment

1-year reservations on Dedicated SQL save ~37%. 3-year saves ~65%. Right for the steady production data warehouse.

Use Spark auto-pause

Variable, often 50%+

Spark pools auto-pause after configurable idle (default 15 min). Aggressive auto-pause prevents idle billing. Right for batch workloads that don't need warm-cluster latency.

FAQ

When should I use Synapse vs Databricks?

Synapse for data warehouse-led workloads (T-SQL, BI, traditional DW patterns). Databricks for Spark-led workloads (data engineering, ML, complex transformations). Both can do both; the question is which has better tooling for your primary use case. Some orgs use both for different teams.

Is Serverless SQL really $5/TB?

Yes, very competitive with Athena ($5/TB). For ad-hoc analysis over data lake (Delta Lake, Parquet on ADLS), Serverless SQL is often the cheapest option in Synapse. Optimize with partitioning and columnar formats to scan less data.

Can I pause Spark pools like SQL pools?

Spark pools auto-pause when idle (no explicit pause needed). When a job runs, the pool resumes (1-5 minutes spin-up). Configure auto_pause_delay_in_minutes to control idle timeout. Aggressive timeouts (5-10 min) save more; longer (30+ min) keep clusters warm for repeated jobs.

What's the smallest viable Synapse setup?

Workspace + Serverless SQL only = ~$0 base + per-query fees. Right for exploration and small workloads. Adding Dedicated SQL DW100c jumps to $876/month minimum. Plan whether you actually need Dedicated SQL before provisioning.