AWS Aurora Serverless Gets a 45% Faster Scaling Engine and Up to 30% Higher Throughput in Platform Version 4
Amazon Aurora Serverless platform version 4 scales 45% faster and delivers up to 30% better throughput at no extra cost, targeting agentic AI workloads.
Overview
Amazon Web Services on April 20, 2026 shipped platform version 4 of Aurora Serverless, its on-demand, auto-scaling configuration for the Aurora relational database service, delivering up to 30% better performance and a 45% improvement in scaling speed, according to the AWS Database Blog. The update is available at no additional cost, and all new Aurora clusters automatically launch on the new platform.
What Changed
Platform version 4 introduces what AWS describes as an enhanced scaling algorithm that takes “additional metrics as signals for scaling decisions,” enabling the service to “intelligently respond to resource competition among concurrent tasks, such as busy web applications and reporting services,” according to the AWS Database Blog.
The most concrete change is in maximum scaling speed. Scaling from 0.5 Aurora Capacity Units (ACUs) to 256 ACUs now takes 22 minutes, compared to 40 minutes on the previous generation — a 45% reduction in ramp-up time, as AWS authors Jiaming Yan, Ashok Kurakula, and Nashad Safa detailed in the AWS Database Blog. AWS also separately doubled the default scaling rate across all serverless clusters and platform versions, requiring no configuration changes from existing customers.
Benchmark Results
AWS published two sets of benchmarks to substantiate the claims.
HammerDB TPROC-C, run at 1,024 virtual users on a 128 ACU configuration, showed that “platform version 4 delivers 27-34% higher NOPM compared to platform version 3” for both Aurora MySQL and Aurora PostgreSQL, according to the AWS Database Blog.
Sysbench read-only benchmark, run with the oltp_read_only.lua script across 250 tables, 16 GB of data, 50 million queries, and 512 threads, showed the platform version 4 cluster completing the workload in 27 minutes using 109 ACUs total, versus 37 minutes and 151 ACUs for version 3. That translates to 27% faster completion with 28% lower cost than platform version 3, and 41% faster completion with 42% lower cost than platform version 2, according to the AWS Database Blog.
Sysbench write-only benchmark, run with oltp_write_only.lua across 50 million transactions and 256 threads, showed platform version 4 achieving 32.9% faster completion with 5.1% lower ACU consumption than version 3, per the AWS Database Blog.
Agentic AI as the Stated Audience
AWS is positioning the update as particularly relevant for AI agent infrastructure. The announcement describes Aurora Serverless as well-suited for “agentic AI applications, which typically have bursts of activity, long idle windows, and unpredictable patterns,” and says the service “handles all of it automatically, scaling capacity with your agents,” according to the AWS Database Blog. The framing aligns with a broader cloud industry pivot toward workloads driven by LLM-based agents rather than human users with more predictable traffic curves.
Industry Reaction
InfoQ noted the improvements come at no extra charge. Corey Quinn, chief cloud economist at The Duckbill Group, offered a characteristically dry assessment: “Aurora Serverless scaling up 45% faster and down to zero, which is coincidentally where my enthusiasm for ‘serverless’ databases that still bill in ACU fractions tends to land.”
Pini Dibask, principal database solutions architect at AWS, highlighted a compounding factor: “What makes this especially interesting (…) is that Database Savings Plans which was announced at re:Invent 2025 offers Aurora Serverless the highest discount of any AWS database service (35% discount),” according to InfoQ.
Upgrade Path
All new Aurora clusters, database restores, and clones automatically launch on platform version 4. Existing clusters on platform versions 1 through 3 can upgrade via pending maintenance action, by stopping and restarting the cluster, or through blue/green deployments, according to the AWS Database Blog.
What We Don’t Know
AWS has not disclosed what specific runtime changes underlie the performance improvements beyond the description of an enhanced scaling algorithm. The benchmarks cover HammerDB TPROC-C and Sysbench workloads but do not include real-world or mixed-workload data. Platform version 3 launched in August 2025, meaning version 4 arrives roughly eight months after the prior generation; AWS has not indicated a cadence for future versions.