Data Clean Room Architecture: 12 Providers Compared (2026)

A data clean room is a controlled environment where two or more parties run approved queries over combined data without either side seeing the other's records. Architectures differ on four axes: where the data sits, who runs the compute, which privacy technique limits the output, and which protocol connects the room to partners. Twelve options are compared here.

1. Where the data sits. Warehouse-native rooms (Snowflake, Databricks, AWS Clean Rooms) query data where it already lives. Walled-garden rooms (Google Ads Data Hub, Amazon Marketing Cloud) hold the platform's event data and let you bring yours in. Vendor-hosted rooms (LiveRamp, InfoSum, Optable, Decentriq) run the join on the vendor's infrastructure, sometimes with the data staying in each party's own cloud account.

2. Who runs the compute. In AWS Clean Rooms the "member who can query" runs the job and a configurable payer is billed (AWS Clean Rooms pricing). In Snowflake the analysis runner pays (Snowflake docs). In enclave rooms the compute is a trusted execution environment that neither party nor the cloud operator can inspect (Microsoft Learn).

3. Which privacy technique limits the output. Allowed-column lists, query approval, minimum aggregation thresholds, noise injection or formal differential privacy, and cryptographic matching so that no party holds the other's cleartext identifier. AWS offers all of these as configuration; its differential privacy went GA on 10 April 2024 (AWS What's New, 10 April 2024) and reached the Spark engine on 16 July 2025 (AWS doc history).

4. Which protocol connects the room to partners. A room that joins your data with one partner's is a project; a room that can produce PAIR IDs or accept ADMaP requests is infrastructure.

"n/d" means the vendor publishes nothing on that cell as of 17 October 2026; "vendor claim" marks figures from the vendor's own site.

The differences are structural.

Membership. AWS requires every member to hold an AWS account, although since 1 December 2024 a member's data can live in Snowflake or Athena (AWS doc history). Snowflake requires every party to be on Snowflake, with Enterprise Edition for data providers, but clouds and regions can differ (Snowflake docs).

Privacy controls. AWS is more prescriptive: analysis rules are first-class objects, minimum aggregation thresholds and comparison controls can now be set inside custom rules (13 August 2026), and differential privacy is a managed feature (AWS doc history). Snowflake controls exposure through the data offering specification and approved templates; its differential privacy capability came from LeapYear and was in preview at GA in March 2024 (InfoWorld, 28 March 2024).

Machine learning. AWS Clean Rooms ML has shipped lookalike modeling (3 April 2024), custom models (7 November 2024), incremental and distributed training (1 July 2025) and synthetic datasets (30 November 2025) (AWS doc history). Snowflake rooms run approved Python code specs (Snowflake docs).

Cost. AWS publishes worked examples: $3,504 a year for a daily three-minute Spark SQL query on 32 CRPUs, and $7,000 a week for a lookalike workflow of 50 million training profiles and ten 2-million-profile segments (AWS Clean Rooms pricing). Snowflake's application is free; you pay the warehouse that runs the analysis (InfoWorld, 28 March 2024).

Lock-in. Both keep data in place. Both also assume the counterparty is on the same platform. Databricks is the one warehouse vendor that markets collaboration with non-Databricks partners through federated sharing (Databricks, 9 June 2025).

PAIR (Publisher Advertiser Identity Reconciliation). PAIR lets an advertiser and a publisher match their first-party identifiers without either side learning the other's list and without a third-party cookie. It uses commutative encryption: each party encrypts its hashed identifiers with its own key, the ciphertexts are exchanged and encrypted again with the other party's key, and the doubly encrypted values can be compared because the order of encryption does not change the result. The matched, encrypted IDs are what the publisher's ad server or SSP can act on. IAB Tech Lab says the protocol "was originally developed by Google Ads team," finalized v1.0 in January 2025 under its Rearc Addressability and PETs Working Group, has since finalized v1.1 with clarifications and an Open PAIR module for Prebid, and has retired the earlier OPJA specification because both solved the same problem (IAB Tech Lab, PAIR, accessed 17 October 2026). Reference cryptographic libraries in Java (Bouncy Castle elliptic-curve operations) and Python (pairid, Curve25519 and Ristretto) are listed in the IABTechLab PAIR repository (IABTechLab/pair on GitHub, accessed 17 October 2026).

ADMaP (Attribution Data Matching Protocol). ADMaP is the attribution counterpart: a privacy-safe way for an advertiser and a publisher or platform to match exposure and conversion records so that attribution can be computed without exchanging cleartext identifiers. IAB Tech Lab published v1.0 Final in February 2025 and moved the specification to GitHub in September 2025 (IAB Tech Lab, ADMaP v1.0, accessed 17 October 2026; InteractiveAdvertisingBureau/ADMAP on GitHub). A webinar replay on "Privacy Safe Secure Attribution with ADMaP" was published on 10 April 2025 (IAB Tech Lab, 10 April 2025).

What the bidstream sees. A clean-room-matched identifier reaches a DSP through the OpenRTB user.eids array. The 2.6-202409 snapshot of 23 September 2024 added three attributes to the EID object for exactly this purpose: inserter (the domain of the entity that added the ID), matcher (the technology that performed the match) and mm (the match method) (openrtb2.x releases; OpenRTB 2.6, section 3.2.27). The 2.6-202505 snapshot of 20 May 2025 added cids to the Data object for content IDs (openrtb2.x releases). An annotated example:

{
  "user": {
    "eids": [
      {
        "source": "pair-protocol.iabtechlab.com",
        "inserter": "publisher.example",
        "matcher": "cleanroom.example",
        "mm": 1,
        "uids": [
          { "id": "BASE64_DOUBLY_ENCRYPTED_PAIR_ID", "atype": 3 }
        ]
      }
    ]
  }
}

Field notes: source identifies the ID namespace; inserter and matcher tell the buyer who put the ID in the request and who did the match; mm is the match-method enumeration from the AdCOM list; atype 3 is a "person-based" agent type. The exact source value for PAIR IDs should be taken from the PAIR v1.1 specification, not from this example. For where this sits in the request lifecycle, see the RTB architecture hub.

Agentic enrichment is the alternative path. IAB Tech Lab's ARTF runs a partner's container inside the host's perimeter instead of exporting from a clean room; Adform and Adsquare shipped the first public integration on 27 May 2026 (ExchangeWire, 27 May 2026). See AAMP vs AdCP.

Both are free at the point of use and both are one-directional: the platform's data never leaves, and your own signals, once uploaded to AMC, cannot be exported (Amazon Ads, accessed 17 October 2026). Treat them as measurement surfaces for spend on those platforms. A network reporting to brands across Amazon, Google and its own channels needs one of the other options as the hub; see Retail media measurement standards explained.

This section is neutral. It lists what a clean room must provide, as a requirements checklist, for teams that cannot or will not place regulated data in a hyperscaler or vendor room: a US bank under GLBA, a pharmacy under HIPAA and state health-privacy laws, or an EU retailer whose counsel has read the CLOUD Act (18 U.S.C. § 2713) and concluded that a Frankfurt region run by a US company does not settle the question (Cornell LII, 18 U.S.C. § 2713). See Data sovereignty defines the future of ad tech and the glossary entries for data residency and the CLOUD Act.

Minimum requirements, derived from what the managed rooms above expose as features:

1. Isolated compute with attested execution. Either physical isolation under your control or a TEE with remote attestation, so a partner can verify what code runs on its data (the model Azure and Decentriq use).
2. A matching layer that never sees cleartext on both sides. Implement PAIR (commutative encryption) for activation matches and ADMaP for attribution matches from the IAB Tech Lab specifications; reuse the reference libraries (IABTechLab/pair; InteractiveAdvertisingBureau/ADMAP). Private set intersection is the fallback for partners that support neither.
3. Analysis rules as data, not as code review. Allowed columns, allowed joins, allowed aggregations, minimum group sizes (the hyperscalers settle between 10 and 50; pick and document yours), and comparison controls on which columns can be compared to literals.
4. A query approval workflow with a tamper-evident log. Every data owner approves every template that touches its tables; every run is logged with template hash, parameters, row counts and runner. Partners will ask for exportable, redacted execution logs, which AWS added on 11 August 2026 (AWS doc history).
5. Output controls. Row suppression below the threshold, optional noise or differential privacy with a per-table budget (AWS's "data access budgets" of 1 October 2025 are the managed equivalent), and an agreed result destination.
6. Identity and entity resolution. Normalisation and hashing rules for email, phone and address; deterministic match first; a documented probabilistic fallback with disclosed error, as the IAB/MRC guidelines require (IAB/MRC, January 2024).
7. Multi-party membership and change control. Adding a member, changing abilities or changing the payer requires approval from every existing member (AWS's change-request model since 18 December 2025).
8. Residency and key custody. Data, keys and logs in the jurisdiction the regulator expects; key material held by the data owner, not the operator.
9. Egress and retention. Hard limits on what leaves (aggregates, PAIR IDs, model scores), retention periods per table, and deletion that a partner can verify.
10. Cost and capacity planning. A managed room bills per query; a self-hosted room bills in hardware and the engineers who maintain the crypto, the approval UI and the audit trail.

• Where are the partners? List the five partners you must match with in the next 12 months and which room each uses; the answer is usually the room with the most overlap plus a protocol bridge for the rest.
• Which protocol do they speak? Ask each vendor, in writing, whether it produces PAIR v1.1 IDs and accepts ADMaP v1.0 requests; no vendor page we reviewed stated either as of 17 October 2026.
• What leaves the room? Aggregates only, or row-level encrypted IDs for activation? The second is more useful and more regulated.
• What is the threshold and who sets it? Minimum group size, and whether it applies per query, per table or per collaboration.
• Who pays and what is the unit? CRPU-hours, warehouse credits, per-profile fees or a license; run your own volumes through a worked example like AWS's.
• Can you leave? Export of your tables, templates and logs at termination.
• Where does it run? Region, cloud, operator nationality, and exposure to the CLOUD Act or an equivalent.
• Is personal data allowed? Azure's preview says no (Microsoft Learn); check the terms.
• Does the output meet the measurement standard you report against? IAB/MRC requires disclosure of attribution basis, windows and extrapolation (IAB/MRC, January 2024); IAB Europe V2.1 requires disclosure of the delivery tool (IAB Europe, V2.1).

For how these rooms are bundled by retail media vendors, see Retail media network platforms compared; for the non-retail networks that depend on them, see the commerce media networks directory.

Clean rooms recur across this site because they are the only measurement layer that works when records cannot move. The non-retail networks in the commerce media networks directory use them for offsite overlap and incrementality; HIPAA-compliant programmatic advertising and The pixel problem in healthcare marketing rely on them to keep health data inside the covered entity; GLBA and advertising with bank data and How a bank can launch a media network without moving customer data place them at the partner boundary; EU sovereign ad tech adds the residency and control tests; and Identity after Privacy Sandbox explains why matched identifiers, not cookies, feed them.

What is a data clean room in simple terms? It is a locked room where two companies put their customer lists on the table, a referee matches them without reading either aloud, and only the totals come out. Data stays under each owner's control, a cryptographic or isolated-compute step performs the join, and output rules suppress anything small enough to identify a person. IAB and IAB Europe published plain-language guides on 29 September 2026 (IAB Europe, 29 September 2026).

Snowflake clean room vs AWS Clean Rooms: what is the difference? AWS Clean Rooms is a managed service with explicit analysis rules, managed differential privacy and a published per-CRPU-hour price; all members need AWS accounts, though data can be in Snowflake or Athena (AWS pricing; AWS doc history). Snowflake Data Clean Rooms is a free native app where the analysis runner pays compute; all parties need Snowflake, Enterprise Edition for data providers, but clouds and regions can differ (Snowflake docs; InfoWorld, 28 March 2024). Pick by where your partners already are.

What is PAIR and do I need it? PAIR is IAB Tech Lab's standard for matching an advertiser's and a publisher's first-party identifiers with commutative encryption so neither sees the other's list; v1.0 was finalized in January 2025 and v1.1 adds a Prebid module (IAB Tech Lab, PAIR). You need it if you want to activate a match on a publisher's inventory through programmatic rather than only measure an overlap. You need ADMaP, published February 2025, if you want attribution across the two sides (IAB Tech Lab, ADMaP).

Are data clean rooms GDPR compliant? A clean room is a tool, not a legal basis. Pseudonymised data is still personal data under GDPR, so you need a lawful basis, a controller or processor agreement with the operator, and, if the operator or its parent is outside the EU, a Chapter V transfer mechanism. The room helps with minimization and purpose limitation; it does not replace consent or legitimate interest. Vendors that keep data in each party's own environment (InfoSum) and EU-based enclave vendors (Decentriq, Zürich) are often chosen for that reason (InfoSum; Decentriq).

How much does a data clean room cost? AWS publishes examples: about $3,504 a year for one three-minute Spark SQL query a day on 32 CRPUs at $2.00 per CRPU-hour, and $21.33 per run for a monthly PySpark measurement job (AWS Clean Rooms pricing, accessed 17 October 2026). Snowflake's app is free plus compute (InfoWorld, 28 March 2024). AMC is free to eligible advertisers (Amazon Ads). Vendor-hosted rooms do not publish prices.

Use the identity and privacy guide to compare the related operating models, definitions and evidence before making this decision.

1. AWS, "Document history for the AWS Clean Rooms User Guide" (entries 12 January 2023 to 6 October 2026; accessed 17 October 2026). https://docs.aws.amazon.com/clean-rooms/latest/userguide/doc-history.html
2. AWS, "AWS Clean Rooms pricing" (accessed 17 October 2026). https://aws.amazon.com/clean-rooms/pricing/
3. AWS, "AWS Clean Rooms Differential Privacy is now generally available" (10 April 2024). https://aws.amazon.com/about-aws/whats-new/2024/04/aws-clean-room-differential-privacy
4. Amazon Ads, "Amazon Marketing Cloud" (accessed 17 October 2026). https://advertising.amazon.com/solutions/products/amazon-marketing-cloud
5. Snowflake, "Snowflake to acquire Samooha" (18 December 2023; closing note 20 December 2023). https://www.snowflake.com/en/blog/snowflake-to-acquire-samooha/
6. InfoWorld, "Snowflake's Data Clean Room promises to ease analysis of PII data" (28 March 2024). https://www.infoworld.com/article/2336670/snowflakes-data-clean-room-promises-to-ease-analysis-of-pii-data.html
7. Snowflake documentation, "Snowflake Data Clean Rooms: introduction" (accessed 17 October 2026). https://docs.snowflake.com/en/user-guide/cleanrooms/introduction
8. Snowflake, "Data Clean Rooms" product page (accessed 17 October 2026). https://www.snowflake.com/en/product/features/data-clean-rooms/
9. Databricks, "Databricks Clean Rooms: now generally available on AWS and Azure" (5 February 2025). https://www.databricks.com/blog/databricks-clean-rooms-now-generally-available-aws-and-azure
10. Databricks, "What's new with Databricks Clean Rooms" (9 June 2025). https://www.databricks.com/blog/whats-new-databricks-clean-rooms
11. Google, "Ads Data Hub: introduction" (legacy documentation, accessed 17 October 2026). https://developers.google.com/ads-data-hub/guides/intro
12. Google, "Ads Data Hub: privacy checks" (accessed 17 October 2026). https://developers.google.com/ads-data-hub/guides/privacy-checks
13. Microsoft Learn, "Perform protected multiparty data collaboration on Azure" (Azure Confidential Clean Rooms; doc dated 22 May 2026). https://learn.microsoft.com/en-us/azure/confidential-computing/confidential-clean-rooms
14. LiveRamp, home page (accessed 17 October 2026; habu.com redirects here). https://liveramp.com/
15. InfoSum, home page (accessed 17 October 2026). https://www.infosum.com/
16. Optable, home page (accessed 17 October 2026). https://www.optable.co/
17. Decentriq, home page (accessed 17 October 2026). https://www.decentriq.com/
18. IAB Tech Lab, "PAIR: Publisher Advertiser Identity Reconciliation" (v1.0 January 2025; v1.1; accessed 17 October 2026). https://iabtechlab.com/pair/
19. IABTechLab, "pair" repository (accessed 17 October 2026). https://github.com/IABTechLab/pair
20. IAB Tech Lab, "Attribution Data Matching Protocol (ADMaP) v1.0 Final" (February 2025). https://iabtechlab.com/standards/addressability-and-pets/attribution-data-matching-protocol-admap-v1-0/
21. IAB Tech Lab, "Webinar replay: privacy safe secure attribution with ADMaP" (10 April 2025). https://iabtechlab.com/standards/addressability-and-pets/webinar-replay-privacy-safe-secure-attribution-with-admap/
22. InteractiveAdvertisingBureau, "ADMAP" repository (accessed 17 October 2026). https://github.com/InteractiveAdvertisingBureau/ADMAP/tree/main
23. IAB Tech Lab, openrtb2.x releases (2.6-202409, 23 September 2024; 2.6-202505, 20 May 2025; accessed 17 October 2026). https://github.com/InteractiveAdvertisingBureau/openrtb2.x/releases
24. IAB Tech Lab, OpenRTB 2.6 specification, section 3.2.27 EID (accessed 17 October 2026). https://raw.githubusercontent.com/InteractiveAdvertisingBureau/openrtb2.x/refs/heads/main/2.6.md
25. ExchangeWire, "Adform and Adsquare launch industry-first Agentic Real-Time Framework integration" (27 May 2026). https://www.exchangewire.com/blog/2026/05/27/adform-adsquare-launch-industry-first-agentic-real-time-framework-integration/
26. PPC Land, "IAB Tech Lab's agent registry hits 10 with Amazon and new deployment types" (11 March 2026). https://ppc.land/iab-tech-labs-agent-registry-hits-10-with-amazon-and-new-deployment-types/
27. IAB Europe, "Data clean rooms for retail and commerce media: two guides, one journey" (29 September 2026). https://iabeurope.eu/data-clean-rooms-for-retail-and-commerce-media-two-guides-one-journey/
28. IAB, "Data Clean Rooms in Commerce: Common Use Cases" (29 September 2026). https://www.iab.com/guidelines/data-clean-rooms-in-commerce-common-use-cases/
29. IAB Europe, "Commerce (incl. Retail) Media Measurement Standards V2.1" (page dated 22 January 2026; PDF May 2026). https://iabeurope.eu/knowledge_hub/iab-europes-commerce-incl-retail-media-measurement-standards-v2/
30. IAB/MRC, "Retail Media Measurement Guidelines" (January 2024). https://www.iab.com/wp-content/uploads/2024/01/IAB_Retail_Media_Measurement_Guidelines_January2024.pdf
31. IAB, "Retail Media Advanced Measurement and Data Collaboration" (31 October 2024). https://www.iab.com/guidelines/retail-media-advanced-measurement-and-data-collaboration/
32. Cornell Legal Information Institute, 18 U.S.C. § 2713 (CLOUD Act). https://www.law.cornell.edu/uscode/text/18/2713

• 18 October 2026: aligned the title and heading and applied recorded source-URL corrections where needed. Factual research cutoff remains 17 October 2026.

• 17 October 2026. First edition at this URL, replacing /blog/data-clean-rooms-2025 (301). 12 options compared, 32 sources. Next review 15 January 2027.