Elastic, pay-as-you-go Spark with EMR Serverless

EMR Serverless Spark decouples compute and storage with per-second billing, reducing burst compute costs by 40%+ for elastic workloads such as month-end close and ad hoc analytics.

Benefits

Compute–Storage Separation

Scale compute independently without re-provisioning storage.

Per-Second Billing

Pay only for the resources used, by the second.

Lower Burst Cost

Reduce burst compute costs by 40%+ for spiky workloads.

End-to-end AI pipeline from Spark to model training

EMR integrates Spark feature engineering with PAI large-model training to deliver an end-to-end pipeline—from data preprocessing to Qwen3 fine-tuning.

Benefits

Integrated Feature-to-Training Flow

Link Spark feature engineering directly to PAI training.

End-to-End Automation

Move from preprocessing to Qwen3 fine-tuning in one pipeline.

Faster Iteration

Shorten the cycle from data preparation to model updates.

Fully managed EMR Serverless StarRocks for sub-second analytics

EMR Serverless StarRocks provides a fully managed, vectorized MPP engine for sub-second ad hoc queries, with automatic scale in/out for traffic spikes, high availability without ops overhead, and 30%–70% lower storage cost via compute–storage separation.

Benefits

Sub-Second Ad Hoc Queries

Vectorized MPP delivers interactive performance at scale.

Auto Scaling

Scale in/out automatically to absorb peak traffic.

Fully Managed HA

Reduce ops overhead while maintaining high availability.

Lower Storage Cost

Cut storage cost by 30%–70% with compute–storage separation.