
Serverless compute scales within limits you set and pauses after inactivity. It isn't free, though, and the wake-up delay catches teams off guard.
This guide covers billing components, auto-pause behavior, a cost-estimation method, workload fit, and the operational trade-offs. All figures use East US pricing in USD.
Key Takeaways
- Pausing stops compute charges, but storage and backup charges continue.
- Auto-pause fits single databases with intermittent usage and tolerance for wake-up delay.
- Compare measured serverless usage against a provisioned alternative before switching.
- A low minimum vCore setting doesn't guarantee low cost if the database stays active or resumes often.
Azure SQL Database Serverless: What You Pay For and What Auto-Pause Does
Serverless is a compute tier for single databases in Azure SQL Database, not a separate service. It still runs on infrastructure. Azure simply scales compute with demand and bills per second.
Key facts to confirm against current Microsoft documentation before you deploy:
- It uses the vCore purchasing model, with General Purpose and Hyperscale tiers. Business Critical isn't supported.
- Hardware is standard-series (Gen5).
- You set minimum and maximum vCores. Memory and I/O limits depend on that range, and the default minimum is 0.5 vCores.
- General Purpose serverless is available in most regions worldwide, East US included, with a maximum of 80 vCores there.
The paused state
After the configured inactivity period, the database pauses. Compute cost drops to zero. Storage remains billable. Paused is not the same as idle: a database that is inactive but still online keeps billing compute.
Azure SQL Database Serverless Cost Breakdown
Serverless cost splits into separate buckets. Treat them separately or the estimate will be wrong.
| Component | East US price or rule |
|---|---|
| Compute | $0.0001450 per vCore-second |
| Data storage | $0.115/GB/month (General Purpose bills provisioned storage) |
| PITR backup | Storage equal to max data size is included; overage is $0.10/GB/month (LRS/ZRS) or $0.20 (RA-GRS/RA-GZRS) |
| Long-term retention | $0.025/GB/month (LRS/ZRS) or $0.05 (RA-GRS/RA-GZRS) |
Rates come from the Azure pricing page and vary by region, agreement, and date. Check the Azure Pricing Calculator for your own configuration.
How compute is billed
Each second, Azure bills the greater of CPU used or memory used, subject to minimum floors. Memory normalizes at 3 GB per vCore. Microsoft's formula is vCore price × max(minimum vCores, vCores used, minimum memory GB ÷ 3, memory GB used ÷ 3).
The catch: a 0.5-vCore minimum with 2.1 GB minimum memory bills 0.7 vCores, not 0.5. Low CPU doesn't mean low billed compute. Azure Hybrid Benefit and Reservations discounts don't apply to serverless.
Minimum and maximum vCore settings shape that bill. A higher minimum improves baseline responsiveness but raises the floor you pay whenever the database is online. A higher maximum handles bursts but allows costlier peaks.

Serverless versus provisioned
| Factor | Serverless | Provisioned |
|---|---|---|
| Billing basis | Per second, actual use | Per hour, fixed capacity |
| Scaling | Automatic | Manual |
| Idle charges | Zero compute when paused | Full compute charge |
| Performance | Variable, with resume delay | Consistent |
| Best fit | Intermittent single databases | Steady, higher utilization |
Auto-Pause: How It Saves Money and What It Costs Operationally
The delay and the conditions
The auto-pause delay runs from 15 to 10,080 minutes in one-minute increments. The default is 60. Set it to -1 to disable pausing.
Throughout the delay, session count and user-workload CPU must both be zero. An open session blocks pausing even with no query running. Pausing generally happens 1 to 10 minutes after the criteria are met. You still pay for compute during the delay itself.
Resume behavior
When activity returns, resume takes about one minute. The first connection attempt starts the resume but returns error 40613, meaning the database is unavailable. Your application must retry.

Plan for:
- Connection retry logic and sensible timeouts
- Health checks that don't wake the database needlessly
- Scheduled jobs, monitoring tools, and connection pools that hold sessions open or trigger resumes unexpectedly
What blocks or reverses pausing
- Active geo-replication and failover groups
- Long-term backup retention
- A DNS alias on the logical server
- A SQL Data Sync sync database, or an Elastic Jobs job database
- Login attempts (including monitoring), plus auditing, Query Store, vulnerability scans, and similar operations
When it's a poor fit
Skip auto-pause for latency-sensitive production services, continuously active databases, and systems with strict availability needs. Frequent resumes also erode savings. Judge idle duration, not just whether an idle interval exists.
How to Estimate Monthly Cost in Azure
The formula is simple:
Monthly total = billed compute + storage + backup/ancillary charges − eligible credits
Paused hours count as zero compute, never zero total cost.
Inputs to collect
- Database size and storage growth
- Active hours, plus average and peak vCore use
- Minimum and maximum vCores, and the auto-pause delay
- Number of resume events
- Backup retention needs, region, and purchasing agreement
Validate with real metrics
The app_cpu_billed metric in Azure Monitor reports billed compute in vCore-seconds. Microsoft's billing documentation says to multiply it by the vCore unit price.
Also track these signals:
app_cpu_percentandapp_memory_percent- Database status (Online, Pausing, Paused, Resuming)
In Cost Management, open Cost analysis, use Accumulated costs, and filter service name to Azure SQL Database.
Illustrative calculation (non-binding)
Microsoft's example: General Purpose, 1 minimum and 4 maximum vCores, six-hour auto-pause delay, active the first two hours of a day.
| Period | Billed compute (vCore-seconds) |
|---|---|
| 00:00-01:00 (4 vCores) | 14,400 |
| 01:00-02:00 (memory-driven) | 14,400 |
| 02:00-08:00 (inactive, online) | 21,600 |
| 08:00-24:00 (paused) | 0 |
| Total | 50,400 |
At $0.000145 per vCore-second, that's about $7.31 for the day's compute. Six of the eight billed hours were idle delay. Storage is added on top.

This is a published daily example, not a monthly projection, and real usage will differ.
Compare against provisioned
Model the same database, region, storage, and period under both options. Then find the idle pattern at which serverless wins. Microsoft gives only qualitative guidance here, with no universal break-even threshold, so use your own measurements.
Validation checklist
- Test cold-start latency and retry behavior
- Run scheduled workloads and confirm they don't block pausing
- Check monitoring visibility, since monitors can cause resumes
- Test peak performance at your chosen maximum
- Review actual billing over a representative period before standardizing

Is Serverless Auto-Pause Right for Your Workload?
Good fit: single databases with intermittent, unpredictable, or dev/test usage, where a delayed first request is acceptable.
Be cautious: continuously active production databases, predictably high utilization, latency-sensitive apps, and systems that can't tolerate a failed first request.
Alternatives worth weighing:
- Provisioned compute for consistent demand
- Provisioned elastic pools when multiple databases can share capacity
- Scheduled controls or rightsizing when usage follows a reliable pattern
Whatever compute model you choose, database compute is only part of the Azure bill. Lucidity addresses a different layer: block storage.
It monitors volumes across AWS, Azure, and Google Cloud and autonomously expands and shrinks disks with zero downtime, cutting block-storage costs by up to 70%. That does not change Azure SQL Database serverless compute or auto-pause billing, but it can support estate-wide FinOps reviews alongside this work.
Frequently Asked Questions
How much does Azure serverless SQL Database cost?
It depends on region, tier, compute consumed, min/max capacity, storage, backups, and discounts. In East US, compute is $0.000145 per vCore-second and storage is $0.115/GB/month. Use the Azure Pricing Calculator for your setup.
Is Azure SQL Database serverless?
Azure SQL Database offers a serverless compute tier for eligible single databases. Not every deployment or tier is serverless; Business Critical, for example, isn't supported.
Is Azure SQL Database free?
Generally no. Microsoft's free tier gives each of up to 10 databases per subscription 100,000 vCore-seconds, 32 GB data, and 32 GB backup monthly. It has no SLA, so treat it as dev and proof-of-concept.
How does Azure SQL Database auto-pause work?
After the configured delay with zero sessions and zero user CPU, the database pauses and compute billing stops. Storage charges continue. Resume takes about a minute.
What are the disadvantages of Azure SQL Database serverless?
Resume delay, required connection retries, variable performance, memory and cache effects, and a billing floor while online. Frequent activity can make provisioned cheaper.
How do I calculate Azure SQL Database serverless cost?
Pick region and configuration, estimate billed vCore-seconds, then add storage and backup charges. Account for the delay period and resume frequency. Validate against Azure Cost Management data.