RDS Serverless (Aurora Serverless v2): The Complete 2026 Guide

Aurora Serverless v2 launched in general availability in April 2022. It solves a specific problem with provisioned RDS instances: you choose a fixed instance size (db.r6g.large, db.r5.xlarge, etc.) and pay for that capacity 24 hours a day, 7 days a week, whether your database is processing thousands of queries per second or sitting completely idle at 3am.

For workloads with predictable, stable traffic, that fixed cost is fine. For workloads with significant variation between peak and off-peak — development environments, event-driven applications, workloads with clear day/night patterns — the fixed cost model means paying for headroom that is only used occasionally.

Aurora Serverless v2 replaces the fixed instance model with Aurora Capacity Units (ACUs) that scale automatically within a range you configure. You set a minimum ACU and a maximum ACU. Aurora adjusts the database’s compute capacity in 0.5 ACU increments between those bounds in response to actual load.

You are billed per ACU per second for the capacity that is actively assigned to your database. At peak, you use more ACUs and pay more. During off-hours, you use fewer ACUs and pay less — potentially zero if scale-to-zero is enabled.

See exactly what you’re overpaying in under 60 seconds. Try the Calculator for free →

Key Changes in 2024–2025

Aurora gained scale-to-zero and a higher capacity limit in late 2024; Serverless v1 reached end of life in March 2025. Check current engine and platform support before estimating costs.

1. Scale to Zero: November 20, 2024

Aurora Serverless v2 now supports scaling down to 0 ACUs — true pause-and-resume behavior.

Before this launch, the minimum capacity for Aurora Serverless v2 was 0.5 ACUs, which costs approximately $0.06/hr = $43.80/month continuously — even when no connections were active. This was a major complaint that separated Aurora Serverless v2 from true serverless behavior. With scale-to-zero, a database that has no active connections for a configurable idle period pauses automatically. During the paused state, no ACU charges apply. Storage and other separately metered charges can continue.

When the first connection request arrives, the database resumes automatically. Resume time: approximately 15 seconds. This cold-start latency makes scale-to-zero appropriate for development and testing environments, and for applications where a 15-second reconnect delay on first use is acceptable. It is not appropriate for production applications with strict latency SLOs where users cannot tolerate a 15-second wait.

Version requirements for scale-to-zero: Aurora PostgreSQL 13.15+, 14.12+, 15.7+, or 16.3+. Aurora MySQL 3.08+. If your cluster runs an older version, upgrade the engine version before enabling 0 ACU minimum capacity. Source: AWS official announcement (November 20, 2024).

2. Maximum 256 ACUs: October 3, 2024

Aurora Serverless v2 expanded its maximum capacity from 128 ACUs to 256 ACUs. Source: AWS official announcement (aws.amazon.com/about-aws/whats-new/2024/10/amazon-aurora-serverless-v2-256-acus, October 3, 2024). 256 ACUs provides approximately 512 GiB of memory with proportional CPU and networking — equivalent to very large provisioned instances. This expansion makes Aurora Serverless v2 viable for high-demand production workloads that previously required provisioned instances at the top of the Aurora instance catalog.

3. Aurora Serverless v1 Retired: March 31, 2025

Aurora Serverless v1 reached end of life on March 31, 2025. Review migration results and current engine support when budgeting for an older deployment. Source: AWS Aurora document history.

The key difference between v1 and v2 that frustrated many teams: v1 could scale to zero from its launch, while v2 required a minimum of 0.5 ACUs from its April 2022 GA. The November 2024 scale-to-zero launch for v2 resolved this gap, but teams who migrated from v1 to v2 between 2022 and November 2024 paid the 0.5 ACU minimum for an extended period.

How Aurora Capacity Units (ACUs) Work

Aurora Serverless v2 measures compute capacity in Aurora Capacity Units (ACUs). Each ACU provides approximately 2 GiB of memory with corresponding CPU and networking resources similar to provisioned Aurora instances. The relationship is consistent: 4 ACUs provides approximately 8 GiB of memory, 16 ACUs provides approximately 32 GiB, 128 ACUs provides approximately 256 GiB, 256 ACUs provides approximately 512 GiB. Source: AWS official Aurora documentation.

Aurora Serverless v2 scales in 0.5 ACU increments. This fine-grained scaling means capacity adjustments are smooth — the database does not jump from 2 ACUs to 4 ACUs in one step. It can be at 2.5, 3.0, 3.5 ACUs as load increases. Scaling up happens within seconds in response to increased connection counts, query load, or memory pressure. Scaling down happens more gradually (over several minutes) to avoid thrashing caused by brief load spikes.

Billing is per second for the ACUs actually assigned to the database. If the database runs at exactly 3.5 ACUs for 10 minutes, the charge is 3.5 ACUs x (600 seconds / 3600 seconds per hour) x $0.12/ACU-hour = $0.07 for those 10 minutes. The per-second billing means you pay precisely for what you use — not rounded to the nearest hour as with some legacy AWS services.

Each ACU has approximately 2 GiB of memory: 8 ACUs is about 16 GiB and 16 ACUs about 32 GiB. Equal memory does not prove equal CPU, throughput or latency. There is no db.serverless.small class; benchmark the actual provisioned and serverless configurations before using a price comparison as a savings estimate.

Also read: RDS Reserved Instances: Complete Guide

Aurora Serverless v2 Pricing in 2026

Configuration	ACU Rate/hr	Storage	I/O	Monthly compute only
Aurora Standard (MySQL or PostgreSQL)	$0.12/ACU-hr	$0.10/GB-month	$0.20/million	$0 while paused; other charges continue
Aurora I/O-Optimized	$0.156/ACU-hr	$0.225/GB-month	Included	$0 while paused; other charges continue
At 0.5 ACU minimum (no scale-to-zero)	$0.12/ACU-hr	$0.10/GB-month	$0.20/million	~$43.80/month
At 4 ACU average	$0.12/ACU-hr	$0.10/GB-month	$0.20/million	$350.40/month compute
At 16 ACU average	$0.12/ACU-hr	$0.10/GB-month	$0.20/million	$1,401.60/month compute
At 256 ACU maximum	$0.12/ACU-hr	$0.10/GB-month	$0.20/million	$22,425.60/month compute

All rates: US East (N. Virginia), October 15, 2026. Verify at aws.amazon.com/rds/aurora/pricing — rates change.

Rates checked against the AWS public RDS price list on October 15, 2026. US East (N. Virginia), USD; monthly examples use 730 hours. Prices exclude tax, credits and separately metered services.

Comparing ACU Cost to Equivalent Provisioned Aurora

Aurora Standard r5.large is $0.29/hr ($211.70 for 730 hours), while 8 ACUs continuously cost $700.80. Both have roughly 16 GiB memory, but this is a price illustration, not a performance-equivalence claim.

For a modeled 176 hours at 8 ACUs and 554 hours at 0.5 ACU, serverless compute is $168.96 + $33.24 = $202.20, $9.50 below continuously running r5.large. If those 554 hours are actually paused, the idealized compute bill is $168.96, a $42.74 difference. Add actual pause/resume time and maintenance; a scheduled provisioned instance might also avoid some idle hours.

For a separate model with exactly half the hours at 16 ACUs and half at 2, the average is 9 ACUs and compute is $788.40. Sixteen ACUs has roughly 32 GiB memory, closer to r5.xlarge (32 GiB, $0.58/hr, $423.40/month) than r5.2xlarge (64 GiB). The former example’s $58 saving used the larger provisioned instance without establishing that it was necessary.

Calculate break-even as the measured comparable provisioned hourly rate divided by the serverless ACU rate, then include storage, I/O, replicas, proxy and commitments. Against continuously running r5.large at $0.29/hr, the compute-only break-even is about 2.42 average ACUs at $0.12/ACU-hour. There is no universal 60–70%-of-peak rule.

Aurora Standard vs I/O-Optimized: Which to Choose for Serverless v2

Aurora Serverless v2 is available in both Aurora Standard and Aurora I/O-Optimized configurations — the same storage options as provisioned Aurora. The choice matters significantly for I/O-intensive workloads.

Aurora Standard

Storage: $0.10/GB-month. I/O: $0.20 per million requests, billed separately. For workloads with moderate I/O, Standard is cheaper because you only pay for the I/O you consume. A Serverless v2 cluster storing 100 GB with 500 million I/O requests per month: $10 storage + $100 I/O = $110/month in storage and I/O charges.

Aurora I/O-Optimized

I/O-Optimized costs $0.156/ACU-hour plus $0.225/GB-month and includes I/O. At the same measured ACU-hours and stored GB, its extra monthly charge is $0.036 × ACU-hours + $0.125 × average stored GB. It is cheaper when avoided Standard I/O charges exceed that premium. There is no universal 500,000-I/Os-per-instance-hour break-even.

For example, 4 average ACUs over 730 hours and 100 GB create a premium of $0.036 × 2,920 + $0.125 × 100 = $117.62. At $0.20 per million I/Os, break-even is 588.1 million I/Os/month. Use metered I/O counts and a comparable workload; aggregate CloudWatch metrics with their documented period and statistic.

Also read: What Is Amazon ElastiCache?

Features: What Aurora Serverless v2 Supports That v1 Did Not

A major limitation of Aurora Serverless v1 was feature parity with provisioned Aurora. v1 did not support read replicas, Global Database, or several other production features. Aurora Serverless v2 was specifically designed to have full feature parity with provisioned Aurora. Source: AWS official Aurora documentation.

Multi-AZ

Multi-AZ serverless clusters bill writer and reader compute separately. Readers in failover tiers 0 and 1 track writer capacity; readers in tiers 2–15 can scale more independently. Include every billed instance in savings comparisons.

Read Replicas

Serverless v2 instances can serve as Aurora read replicas. A provisioned Aurora cluster can have Serverless v2 read replicas that scale capacity based on read traffic — useful for burst read scenarios where provisioned reader instances would be over-provisioned most of the time. The inverse is also supported: an Aurora cluster with a provisioned writer and Serverless v2 readers, or a fully Serverless v2 cluster with writer and readers all using ACU-based scaling.

Global Database

Aurora Serverless v2 is compatible with Aurora Global Database, enabling a single Aurora cluster to span multiple AWS Regions with sub-second replication lag. Secondary region clusters can use Serverless v2 instances for read scaling in those regions. Source: lushbinary (February 2026) and AWS official Aurora documentation.

Aurora Backtrack

Aurora Backtrack allows you to rewind a database cluster to a specific point in time without restoring from a backup. This is supported for Serverless v2. Backtrack window size is configurable; larger windows incur storage costs.

RDS Proxy

RDS Proxy pools database connections. Its charge is separate: $0.015 per ACU-hour for Serverless v2 in N. Virginia, with an 8-ACU billing minimum; provisioned targets use vCPU pricing. Adding a proxy also prevents automatic pause because it maintains database connections. Include both effects when estimating idle savings.

Performance Insights

Review CloudWatch Database Insights pricing for current monitoring costs. Standard mode includes up to seven days of detailed database and per-query metrics at no additional cost; Advanced mode and other metered monitoring can add charges. Source: AWS Database Insights documentation.

When to Use Aurora Serverless v2 vs Provisioned Aurora + Reserved Instances

This is the decision that determines whether Serverless v2 saves or costs more for your specific workload. The answer depends on three variables: your workload’s peak-to-average utilization ratio, whether your version supports scale-to-zero, and whether your minimum required ACU count makes Serverless v2 economically viable.

Scenario	Recommended Choice	Reason	Approx saving
Dev/test environment, infrequent use, version supports 0 ACU	Serverless v2 (scale-to-zero)	Zero instance compute while actually paused; storage and other charges continue.	Only paused periods have zero instance compute
Variable workload with measured low average ACU consumption	Serverless v2	Scales down during off-hours, saving vs fixed provisioned cost.	Workload-dependent; calculate comparable totals
Steady-state production workload, predictable load, 12+ months stable	Provisioned + Reserved Instances	Sustained ACU cost at high utilization exceeds provisioned RI rate.	Use the exact RI quote and measured serverless cost
Spiky workload: 10x traffic bursts multiple times daily	Serverless v2	Scales in seconds to handle burst. No pre-provisioning for peak.	Avoid over-provisioning cost of sizing for peak
Mixed baseline + burst: high steady baseline with occasional spikes	Provisioned (baseline) + Serverless v2 readers (burst)	RI covers predictable baseline. Serverless handles burst cheaply.	Best of both: RI savings on base + Serverless flexibility for bursts
Microservice with Lambda, unpredictable connections	Serverless v2 + RDS Proxy	Serverless scales with function invocations. Proxy pools connections.	Proxy adds a separate charge and prevents auto-pause.

These are evaluation criteria, not guaranteed savings percentages. Compare the same workload, availability and service requirements using current rates.

The Hidden Costs: What Is Not Included in the ACU Rate

The ACU rate covers compute only. Three additional cost categories apply and are not reduced by the ACU billing rate.

Storage Charges

Aurora storage is billed independently of ACU usage. Under Aurora Standard, storage costs $0.10/GB-month. Aurora auto-scales storage in 10 GB increments as your data grows, up to 128 TB. You are billed for storage consumed, not storage allocated. Even when the Serverless v2 cluster is paused at 0 ACUs, storage charges continue. A cluster storing 50 GB at Aurora Standard: $5/month in storage charges continuously, regardless of whether the database processed zero queries that month.

I/O Charges (Aurora Standard Only)

Under Aurora Standard, each I/O request is charged at $0.20 per million. For OLTP workloads with high read and write rates, I/O charges can exceed ACU charges. A heavily loaded Serverless v2 cluster processing 2 billion I/O requests per month: 2,000 x $0.20 = $400/month in I/O charges alone. Monitor VolumeReadIOPs and VolumeWriteIOPs in CloudWatch to evaluate whether switching to I/O-Optimized reduces total cost. Source: AWS official Aurora pricing page.

Backup Storage

Aurora includes a backup allowance tied to cluster storage; excess backup in this Region costs $0.021/GB-month. If the measured allowance is 100 GB and total backup usage is 500–1,000 GB, chargeable excess is 400–900 GB, or $8.40–$18.90/month. Retention duration alone does not determine the footprint.

RDS Proxy (Optional)

RDS Proxy is $0.015 per ACU-hour in N. Virginia, with an 8-ACU billing minimum. At a steady 8 ACUs for 730 hours, proxy costs $87.60 in addition to $700.80 of Standard database compute, totaling $788.40 before other charges. It does not double database compute charges at this utilization. At low ACUs, the proxy minimum matters. Additional proxy endpoints can incur PrivateLink charges, and proxy connections prevent auto-pause. Sources: RDS Proxy pricing and AWS proxy billing minimum guidance.

Discount Options for Aurora Serverless v2

Database Savings Plans: The Only Commitment Discount for Serverless v2

Reserved Instances do not apply to Aurora Serverless v2. Reserved Instances are specific to provisioned instance types and do not cover ACU-based billing. For Aurora Serverless v2, the only commitment-based discount mechanism is Database Savings Plans.

Database Savings Plans cover eligible Aurora Serverless usage with a one-year commitment. The program maximum is 35%, not a guaranteed discount for every usage type. Eligible usage consumes commitment at discounted rates; usage beyond that hourly commitment remains On-Demand. A plan does not automatically discount an unlimited scale-up.

Illustrative calculation only: steady 8-ACU usage costs $700.80/month On-Demand. If an applicable quote offered a 35% discount, full hourly coverage would require $0.624/hour of commitment and cost $455.52/month, saving $245.28/month or $2,943.36 over 8,760 hours. This is not a verified Aurora offer. Obtain the actual rate, and model each hour: an average of 8 ACUs does not guarantee full commitment utilization.

Usage.ai evaluates Database Savings Plans against actual hourly ACU consumption and scale-to-zero behavior. There is no universal two-ACU or 600-hours threshold that guarantees savings. Compare avoided On-Demand charges with the complete commitment obligation, including unused hours.

How to Configure Aurora Serverless v2: Min, Max, and Scale-to-Zero

Setting Minimum Capacity

The minimum ACU determines the lowest capacity the database will maintain. For production workloads where cold-start latency is unacceptable: set minimum to at least 0.5 ACUs (or higher based on your baseline traffic). This ensures the database always has some compute capacity and can respond to queries without a resume delay. For development and testing environments where cold-start latency is acceptable: set minimum to 0 ACUs (requires supported engine version) to enable scale-to-zero and eliminate compute charges during idle periods.

Setting minimum too high wastes money. Setting minimum too low (0 ACU on a production database without acceptable cold-start handling) creates user-facing latency when the database resumes. The right minimum for production: monitor your P1 CPU utilization during off-peak periods. If the database never drops below 20% CPU at 4 ACUs, your minimum should be 2-4 ACUs to avoid unnecessary scaling cycles.

See exactly what you’re overpaying in under 60 seconds. Try the Calculator for free →

Setting Maximum Capacity

The maximum ACU determines the upper bound of automatic scaling. Setting maximum too low means the database is constrained during traffic spikes and query latency increases. Setting maximum too high does not cost more (you only pay for ACUs actually consumed) but removes cost predictability since an unexpectedly large query or runaway connection pool could temporarily scale to a very high ACU count.

Recommended approach: set maximum to at least 2-4x your expected peak ACU consumption, verified from load testing. For new workloads with no baseline data: start at 16 ACUs maximum, monitor for 30 days, and adjust. CloudWatch metric: ServerlessDatabaseCapacity shows current ACU consumption in real time. Source: AWS official Aurora documentation.

Enabling Scale-to-Zero on Eligible Clusters

For new clusters: in the AWS Console, set minimum capacity to 0 ACU. Verify the selected engine version meets requirements (Aurora PostgreSQL 13.15+, 14.12+, 15.7+, 16.3+, or Aurora MySQL 3.08+). For existing clusters: update the engine version if necessary, then modify the cluster to set minimum capacity to 0.

Configure your application to handle the 15-second resume latency. For applications using connection pools with configurable timeout settings: set the connect timeout to at least 20 seconds. For Lambda functions: configure the connection timeout to exceed the resume duration. AWS recommends configuring client timeout settings such as connectTimeout and sslResponseTimeout to exceed the 15-second resume time. Source: AWS Database Blog (November 2024).

How Usage.ai Handles Aurora Serverless v2 Cost Optimization

Aurora Serverless v2 creates a specific challenge for commitment purchasing: the ACU consumption varies, making it harder to size a Database Savings Plan commitment than a fixed-instance RI purchase. Committing too much wastes money if the workload scales down. Committing too little leaves savings on the table.

Usage.ai analyzes recent hourly ACU consumption to estimate a conservative eligible usage floor. Convert covered usage to the actual discounted Savings Plans rate to size the commitment. Confirm seasonality and future changes; historical minimum usage does not guarantee future utilization.

For Serverless v2 clusters still running at the 0.5 ACU minimum without scale-to-zero enabled, Usage.ai identifies engine version upgrade opportunities that unlock the 0 ACU minimum — particularly valuable for dev and staging clusters. A cluster running at 0.5 ACU minimum 24/7 costs $43.80/month in compute. Enabling scale-to-zero on a cluster that is active 8 hours per day reduces that to approximately $0 during off-hours, with full ACU billing only during active periods.

If a DSP commitment for Serverless v2 becomes underutilized — because a cluster is decommissioned, migrated to provisioned, or the application traffic drops permanently — Usage.ai provides cashback on the unused commitment in real money. Fee: percentage of realized savings only.

See how Usage.ai optimizes Aurora Serverless v2 and Database Savings Plans

 

 

Frequently Asked Questions

1. What is RDS Serverless?

‘RDS Serverless’ refers to Aurora Serverless v2 — the on-demand, auto-scaling configuration of Amazon Aurora, which is an engine of Amazon RDS. There is no standalone product called RDS Serverless. Aurora Serverless v2 scales database capacity in Aurora Capacity Unit (ACU) increments between a configurable minimum (0 for scale-to-zero, supported since November 2024) and maximum (up to 256 ACUs). It supports Aurora MySQL and Aurora PostgreSQL. Aurora Serverless v1 retired March 31, 2025. Source: AWS official Aurora documentation.

 

2. How much does Aurora Serverless v2 cost?

Aurora Serverless v2 charges $0.12 per ACU-hour for Aurora Standard configuration (MySQL or PostgreSQL, US East N. Virginia). Storage: $0.10/GB-month. I/O: $0.20 per million requests. Minimum cost with scale-to-zero enabled: $0 compute during fully idle periods (storage charges continue). At 0.5 ACU minimum (no scale-to-zero): approximately $43.80/month compute. At 4 ACU average: approximately $350/month compute. At 16 ACU average: approximately $1,401/month compute. No Reserved Instances available — Database Savings Plans provide up to 35% savings. Verify at aws.amazon.com/rds/aurora/pricing — rates change.

 

3. Can Aurora Serverless v2 scale to zero?

Yes, since November 20, 2024. Aurora Serverless v2 now supports a minimum capacity of 0 ACUs for supported engine versions: Aurora PostgreSQL 13.15+, 14.12+, 15.7+, 16.3+, and Aurora MySQL 3.08+. With 0 ACU minimum, the database pauses automatically after a configurable idle period (no active connections). Resume time when the first connection arrives: approximately 15 seconds. During the paused state, no compute charges apply — storage and other applicable charges continue. Source: AWS official announcement (November 20, 2024).

 

4. What is an Aurora Capacity Unit (ACU)?

An ACU is the billing unit for Aurora Serverless v2. Each ACU provides approximately 2 GiB of memory with corresponding CPU and networking resources. Aurora Serverless v2 scales in 0.5 ACU increments. 8 ACUs provides approximately 16 GiB of memory — roughly equivalent to a db.r6g.large provisioned instance. Billing is per second: if the database runs at 3.5 ACUs for one hour, the charge is 3.5 x $0.12 = $0.42. Source: AWS official Aurora documentation.

 

5. Do Reserved Instances apply to Aurora Serverless v2?

No. Reserved Instances apply to provisioned Aurora instances (db.r6g.large, db.r7g.xlarge, etc.) but do not cover ACU-based Serverless v2 billing. The commitment discount mechanism for Aurora Serverless v2 is Database Savings Plans, which provide up to 35% savings on a 1-year commitment. DSP applies automatically to eligible ACU usage without being tied to a specific ACU range or cluster. Source: AWS official Aurora pricing and multiple independent analyses consistent with this.

 

6. What is the difference between Aurora Serverless v1 and v2?

Aurora Serverless v1 retired March 31, 2025 and is no longer supported. Key differences: v1 could scale to zero from its initial launch; v2 required a 0.5 ACU minimum until November 2024. v2 supports features v1 did not: read replicas, Global Database, Multi-AZ, Performance Insights, Aurora Backtrack. v2 scales more finely (0.5 ACU increments versus larger v1 scaling steps) and scales more quickly in response to load. v2 has full feature parity with provisioned Aurora. Any cluster still running v1 should have been migrated by March 31, 2025. Source: AWS official Aurora documentation and InfoQ.

 

7. When is Aurora Serverless v2 more expensive than provisioned Aurora?

Compare measured ACU-hours with a provisioned configuration that meets the same performance and availability requirements. At 8 ACUs continuously, Standard serverless compute costs $700.80 over 730 hours; r6g.large at $0.26/hr costs $189.80, but equal memory does not establish equal performance. No universal average-to-peak percentage proves which configuration is cheaper. Add storage, I/O, replicas, proxy and commitment utilization.