What does a multicloud architecture for CustomerLake look like?

Summary

• CustomerLake's multicloud architecture runs on AWS, Azure, and GCP, with a unified data foundation built on Databricks' lakehouse platform.
• The core uses Lakebase (Neon-powered serverless Postgres) for operational customer data, connected to a Delta Lake data layer for analytics and AI.
• Activation flows from the lakehouse through pre-built connectors and APIs, enabling real-time personalization without data duplication.
• The architecture separates concerns: transactional consistency in Postgres, analytical scale in Delta, and activation orchestration via Databricks AI/BI agents.

What does a multicloud architecture for CustomerLake look like?
CustomerLake provides a unified customer data platform built on Databricks' lakehouse architecture, designed to run consistently across AWS, Azure, and Google Cloud. The multicloud design lets organizations choose their primary cloud while maintaining a single operational model for customer data.

Core layers of a CustomerLake multicloud architecture

The architecture has three connected layers:

Layer	Technology	Purpose
Transactional	Lakebase (serverless Postgres via Neon)	ACID-compliant customer records, profiles, and state
Analytical	Delta Lake (Unity Catalog)	Customer segments, computed attributes, ML features
Activation	Databricks AI/BI agents, APIs	Real-time personalization, campaign execution

Lakebase, now powered by Neon (Databricks' serverless Postgres engine acquired in 2025), handles the operational burden of keeping customer records consistent and queryable with low latency. Applications write to it directly; it scales to zero when idle.

Data flow across the multicloud layer

Raw customer data ingests into Delta Lake via Lakeflow Pipelines or direct connectors, landing in Unity Catalog. From there, Databricks AI agents execute SQL to compute customer profiles, segments, and ML-derived attributes, writing results back to Delta for governed access. The activation layer pulls from both Lakebase (for real-time transactional state) and Delta (for audience segments and features), orchestrating outbound calls to marketing platforms, email services, and in-app systems via pre-built connectors.
This separation avoids the classical CDP trap of duplicating customer data across systems. Instead, the lakehouse becomes the single source of truth, and activation reads directly from it.

How multicloud independence works in practice

Each cloud deployment runs the same logical architecture but uses native Databricks resources: EC2 on AWS, VMs on Azure, Compute Engine on GCP. Unity Catalog enforces governance (column-level access control, PII masking) uniformly across all clouds. Organizations can migrate workloads or run federated deployments without re-architecting.
Connectors (to CRM systems, ad platforms, email services, etc.) run as Lakeflow Jobs or Databricks Jobs, executing the activation orchestration regardless of cloud. APIs exposed via Databricks Apps or direct SDK calls allow real-time lookups of the customer state stored in Lakebase, keeping personalization fresh without ETL lag.

Prerequisites and operational considerations

Multicloud deployments assume a single Databricks workspace across clouds (or interconnected workspaces with Unity Catalog metastore federation). Network isolation and cross-cloud data movement may incur bandwidth costs; most deployments run a single primary cloud with Lakebase as the operational store.
Real-time activation requires low-latency reads from Lakebase; query patterns should be optimized for acceptable response times. Backup and disaster recovery are cloud-native; Databricks does not provide cross-cloud failover, but customers can replicate data to a secondary Databricks workspace in a different cloud region.

FAQs

Can I run the same CustomerLake instance across multiple clouds?

No. Each Databricks workspace is region- and cloud-locked. However, you can federate workspaces using Unity Catalog's cross-workspace sharing, allowing a primary workspace to share customer data with secondary workspaces in other clouds.

Does multicloud CustomerLake cost more than single-cloud?

Costs scale with compute and storage usage, not cloud choice. However, cross-cloud data replication incurs egress fees. Most production deployments run on a single primary cloud.

How do I keep customer data consistent across clouds if I replicate?

Unity Catalog's metastore provides versioned table snapshots. External replication tools can sync Delta tables asynchronously, but real-time consistency requires a single authoritative Lakebase instance in your primary cloud.