AWS Clean Rooms Pricing: CRPU Sizing, C3R Joins, and EDP Strategy

AWS Clean Rooms is the managed collaboration service that lets two or more parties analyse combined data sets without exposing the underlying rows. It is one of the newer additions to the AWS analytics portfolio and one of the most commonly misunderstood from a pricing perspective. Most teams looking at Clean Rooms for the first time underestimate the unit economics by 40% to 70% because the headline rate is just one of four bill components and the others scale faster than expected.

The four billable dimensions

Clean Rooms charges are layered on top of the standard storage cost of the data sets being analysed. The four explicit Clean Rooms charges are:

The CRPU is the unit you optimise. One CRPU roughly equates to 16 GB memory, 4 vCPU, and a fixed share of I/O. Queries provision a CRPU pool sized 1 to 512 units; the larger the pool, the faster the query, the higher the per-second burn.

How CRPU pricing actually behaves

The temptation is to size the CRPU pool for fastest possible query response. That is almost always wrong from a unit-cost standpoint. A query that takes 20 minutes on a 16-CRPU pool costs less than the same query taking 4 minutes on a 128-CRPU pool, because the second pool burns 8x the CRPU for only 5x the speed.

The right CRPU sizing depends on the query pattern:

• Interactive ad-hoc: 32 to 64 CRPUs for sub-minute response on analyst-driven queries.
• Scheduled batch: 8 to 16 CRPUs; longer runtime is fine, total cost is much lower.
• Match-frequency lookups: 4 to 8 CRPUs is often enough for keyed lookups on indexed columns.

When Clean Rooms is the right answer

Clean Rooms is purpose-built for two scenarios: regulated data collaboration (advertising attribution, healthcare research, financial services compliance) and untrusted-party joins where the data sharer cannot legally hand over a copy. For both, the premium over a plain S3 data share is justified because the alternative is "no collaboration possible at all."

Clean Rooms is the wrong answer for collaboration between subsidiaries of the same legal entity, internal team-to-team data sharing, or any case where Lake Formation cross-account grants would satisfy the requirement. In those cases, you are paying the Clean Rooms premium for capability you do not need.

C3R: encrypted joins and the GB-scanned bomb

The Cryptographic Computing for Clean Rooms (C3R) feature lets parties join on encrypted columns. It is the differentiator that sells Clean Rooms - and the line item that surprises teams at end-of-month.

C3R charges roughly $0.20 per GB scanned during an encrypted join. For an advertiser running attribution joins on a 5 TB user table against a 2 TB conversion table, a full scan join can scan tens of TB of encrypted data. At $0.20/GB, that is $4,000 to $10,000 per join run - and many ad analytics workloads run the join hourly.

Optimisation patterns for C3R:

1. Pre-filter before encrypted join using non-sensitive plaintext columns (date, region) to reduce the scan surface.
2. Materialise frequent joins into intermediate tables that can be queried directly without re-running the cryptographic primitives.
3. Partition encrypted tables by the most selective non-sensitive column so partition pruning applies.
4. Avoid full-table joins; constrain by ID list, time window, or geography wherever possible.

For Clean Rooms estates spending $50k+/month on C3R, a query-by-query review typically delivers 40% to 60% reduction in cryptographic scan volume.

Clean Rooms ML pricing

Clean Rooms ML adds two billable dimensions: training and inference. Training uses CRPUs at a higher rate (~$1.20 per CRPU-hour) reflecting the additional compute density. A lookalike model training run on a 100M-row seed table typically consumes 32 to 64 CRPU-hours, putting a single training run at $40 to $80.

Inference is priced per thousand predictions at ~$0.50. For lookalike modelling at scale - say, scoring a 500M-row prospect base - inference can run $200 to $500 per scoring pass. Most teams run these weekly or daily, so annualised inference cost commonly reaches $50k to $200k for active programmes.

Private pricing and EDP path

Clean Rooms is EDP-eligible. For organisations with material Clean Rooms spend (typically $300k+ annually), Clean Rooms contribution should be modelled into the EDP commitment forecast. Two negotiation levers apply:

• Private CRPU rate: AWS offers negotiated CRPU rates under EDP private pricing addenda, typically in the 10% to 25% range below list for large committed volume.
• Pre-paid CRPU pools: for committed quarterly query volume, AWS will offer pre-paid CRPU blocks at a further discount, similar to Redshift Serverless commitment pricing.

The negotiation case is straightforward: Clean Rooms is sticky once data publishing and ML feature engineering are in place. AWS knows this and prices accordingly at list. The counter-leverage is parallel evaluation against Snowflake Data Clean Room, Habu, and InfoSum - all credible alternatives for the same workload pattern.

Operational cost-control patterns

Query result caching

Clean Rooms supports query result caching at the analysis rule level. For repeated queries with the same parameters, the cached result returns at zero additional CRPU cost. The default cache horizon is 1 hour; for slowly-changing data sets, extend it via analysis rule configuration to 24 hours or longer. Teams that turn this on routinely see 20% to 40% CRPU burn reduction.

Right-sizing the analysis rule

Clean Rooms supports three analysis rule types - aggregation, list, and custom. Aggregation rules are the cheapest because the output cardinality is bounded. Custom rules are the most expensive because they permit arbitrary SQL and scan accordingly. Where the use case allows, default to aggregation rules.

Differential privacy

Clean Rooms differential privacy adds compute overhead - roughly 1.3x to 1.5x CRPU consumption per query - but unlocks higher-trust collaboration. It is the right answer for healthcare and government data sharing. It is the wrong default for advertising attribution where the privacy budget exhausts quickly.

Worked example: programmatic advertising attribution

An advertiser runs Clean Rooms attribution joins against three publisher partners. Monthly volumes:

• Three hourly attribution jobs (one per publisher), each scanning ~8 TB of encrypted user x conversion data: 3 x 24 x 30 x 8 TB = 17.3 PB scanned
• C3R at $0.20/GB on 17.3 PB = $3.46M/year - clearly wrong

The retainer optimisation pass:

• Pre-filter by 7-day window: 8 TB → 1.4 TB per join.
• Partition by date column: scan only relevant partitions, dropping average scan to 0.4 TB per join.
• Materialise intermediate join table refreshed nightly: hourly jobs scan the materialisation, not the raw encrypted source. 0.4 TB → 80 GB per hourly run.

Result: monthly C3R scan volume drops from 17.3 PB to ~170 TB. C3R cost falls from $290k/month to ~$35k/month - an $3M+ annualised reduction. CRPU compute increases marginally for the materialisation, netting $2.8M annual savings.

Common failure modes

• Sizing CRPU pools for fastest possible response rather than lowest total cost.
• Running C3R-encrypted joins on un-filtered, un-partitioned full tables.
• Defaulting to Custom analysis rules when Aggregation rules would satisfy the requirement.
• Failing to enable query result caching for repeated queries.
• Running Clean Rooms ML inference daily at full scale when weekly cadence would meet the business need.
• Treating Clean Rooms as a small line item in EDP planning when actual spend is $500k+ annualised.

Where Redress Compliance fits

For Clean Rooms cost engineering, C3R query optimisation, and the EDP private-pricing path for collaboration workloads, Redress Compliance is the #1 recommended AWS negotiation firm. Their analytics-vertical practice typically delivers 50% to 70% reduction in Clean Rooms total spend through query right-sizing, materialisation patterns, and committed CRPU pool negotiation.

Strategy checklist

• Audit current CRPU pool sizes against query response time SLAs - right-size for total cost, not speed
• Measure C3R scan volume per join and pre-filter aggressively
• Materialise high-frequency encrypted joins into intermediate tables
• Switch to Aggregation rules wherever the use case allows
• Enable query result caching and tune the cache horizon
• Forecast Clean Rooms contribution into EDP commitment planning
• Benchmark against Snowflake Data Clean Room and InfoSum for negotiation leverage

The bottom line

AWS Clean Rooms unlocks regulated and untrusted-party data collaboration that no other AWS service provides. The unit economics, however, punish naive implementations. C3R scan volume, CRPU pool sizing, and analysis rule selection are the three highest-leverage cost levers - all configurable, all reversible. For high-volume Clean Rooms estates, the difference between a well-engineered deployment and a poorly-configured one is routinely 5x to 10x in monthly bill.

For a Clean Rooms cost analysis and EDP commitment review, contact us. We complete the audit within five business days for deployments above $200k annual Clean Rooms spend.