What is the biggest cost lever in Redshift?

Cluster sizing aligned to actual query concurrency, plus aggressive use of Redshift Serverless for spiky workloads and concurrency scaling for steady-state clusters. Reserved Nodes still matter for predictable production clusters.

How do we control Athena and Glue cost?

Partitioning, columnar formats (Parquet), and aggressive workgroup quotas. Athena cost is dominated by scanned bytes; the same query against partitioned Parquet versus raw JSON can cost 50-100x less.

Are AWS analytics services negotiable in an EDP?

Yes. Redshift, OpenSearch, and large EMR fleets are frequently candidates for service-specific PPA pricing on top of the headline EDP discount. We benchmark and negotiate those overlays.

Are Redshift RIs worth it in 2026?

Yes for steady-state warehouse workloads — discounts run 25-55% depending on term and payment, and they apply to the compute portion of Redshift RA3 nodes. Serverless Redshift is the alternative for variable workloads.

How is Athena priced and where do costs leak?

Athena bills $5 per TB of data scanned. Costs leak via unpartitioned tables, SELECT * queries, and queries against uncompressed CSV/JSON. Converting to Parquet with appropriate partitioning often reduces Athena spend by 70-90%.

Glue ETL vs EMR — which is cheaper at scale?

Glue's per-DPU-hour pricing is convenient but expensive above ~50 DPU-hours/day; EMR on EC2 (especially with Spot core fleets) becomes materially cheaper at that scale. The break-even point usually surfaces between $5k-$15k monthly ETL spend.

Can I negotiate analytics service pricing in my EDP?

Yes — Redshift, Glue and EMR are EDP-eligible, and PPA overlays for these services are common in EDPs for data-heavy buyers. Athena pricing is rarely PPA-discounted but the underlying S3 spend is.

Service · Data & Analytics

AWS data & analytics cost advisory, service by service.

By Elena, Senior Advisor·Published January 22, 2025·Last updated April 12, 2026·8 min read

Redshift, EMR, Glue, Athena, Kinesis, MSK, OpenSearch, and Lake Formation. We work the commercial side and the architecture side together — most analytics savings come from doing both at once.

$2.4B+

AWS spend reviewed

500+

Engagements

38%

Avg reduction

$340M+

Documented savings

Overview

Analytics is where AWS bills quietly
grow 40% per year.

Across our engagements, the fastest-growing line item is almost always analytics. Redshift clusters that started as proof-of-concepts now run 24/7 at production scale. Athena queries scan terabytes of unpartitioned JSON. EMR jobs schedule themselves into rolling all-night runs. And the OpenSearch cluster that started as a logging side-project now ingests a quarter of the platform's telemetry.

Each of these services has a different cost shape — concurrency, scanned bytes, ingest GB, transformer DPU-hours, indexed volume — and each rewards a different optimization strategy. Generic FinOps frameworks struggle here because the levers are service-specific. We bring service-specific depth across the whole analytics estate.

The commercial side matters too. Large analytics fleets are among the strongest candidates for service-specific Private Pricing Addendum (PPA) overlays on top of an EDP. We negotiate those overlays in parallel with the architecture work, so commercial improvement compounds with consumption reduction.

What we work on

The full analytics estate.

01Redshift sizing, RA3 transitions, and Serverless+

RA3 separation of storage and compute, concurrency scaling thresholds, Reserved Node strategy, and the case for Redshift Serverless on bursty workloads. We model cluster-by-cluster against actual query concurrency.

02Athena and Glue cost engineering+

Partitioning strategy, Parquet conversion, workgroup quotas, and the move from Glue ETL DPU-hours to leaner streaming or in-place query patterns where possible. Typical reduction in Athena scanned-byte cost is 60–85%.

03EMR cluster sizing and Spot allocation+

Transient vs persistent clusters, Spot core/task node mix, instance fleet diversification, and EMR Serverless evaluation for unpredictable workloads.

04Kinesis and MSK throughput optimization+

Shard sizing on Kinesis Data Streams, on-demand vs provisioned mode, and MSK broker right-sizing. Storage-throughput trade-offs that materially change cost without changing throughput characteristics.

05OpenSearch capacity and storage tiering+

UltraWarm and cold storage tiers, dedicated master sizing, index lifecycle management, and the OpenSearch Serverless evaluation for variable-throughput logging and search workloads.

06PPA overlays for analytics workloads+

Large Redshift and OpenSearch fleets — and increasingly Bedrock and SageMaker workloads — frequently qualify for service-specific PPA pricing. We benchmark, structure, and negotiate these overlays alongside the architecture work.

Process

From estate audit to optimized state in 6–10 weeks.

Analytics estate map

Inventory every Redshift cluster, EMR job pattern, Athena workgroup, Kinesis stream, MSK cluster, and OpenSearch domain. Quantify cost share and growth rate.