Salesforce Data Cloud

Data Cloud is a Decoupled, Zero-Copy Lakehouse

If Traditional Core Salesforce is Transactional and Row-Based, and External Data Warehouses are Analytical but Disconnected — Data Cloud is a Decoupled, Zero-Copy Lakehouse.

Architectural Breakdown: CRM DB vs. Warehouse vs. Data Cloud

Traditional Salesforce (Core CRM): Transactional & Relational

• The Brain: Lives in the Oracle/Postgres relational database engine.
• State: Highly structured, strictly governed by object schemas (Accounts, Contacts).
• The Behavior: Engineered for OLTP (Online Transaction Processing). It is incredibly fast at finding one specific record and updating it. It fails miserably at aggregating 50 million web-click events.

Traditional Data Warehouse (Snowflake/Redshift): Analytical & Heavy

• The Brain: Massive parallel processing servers.
• State: Data is extracted, transformed, and loaded (ETL) in massive batches.
• The Behavior: Great for OLAP (Online Analytical Processing) and historical reporting, but completely disconnected from the Salesforce UI and business triggers.

Salesforce Data Cloud: Decoupled Lakehouse

• The Brain: Lives on Hyperforce (AWS infrastructure), utilizing big-data query engines like Trino/Presto and Vector databases.
• State: Handles both batch and real-time streaming data. It doesn't force a strict schema on ingest.
• The Behavior: It ingests massive volumes of raw data, maps it to a unified semantic model, resolves identities, and then exposes that data back to Core Salesforce without actually copying the billions of rows into the CRM database.

Conceptual Architecture Summary

• Lakehouse Architecture: It combines the flexibility and cheap storage of a data lake (storing raw JSON, CSVs, unstructured text) with the management and querying features of a data warehouse.
• Zero-Copy Federation: Instead of building complex ETL pipelines to copy data from Snowflake/BigQuery into Salesforce, Data Cloud creates a "live mount." It queries the external database in real-time.
• Identity Resolution: The engine mathematically evaluates fragmented records (e.g., an email from Marketing Cloud, a device ID from a website, a phone number from Service Cloud) and merges them into a single Unified Individual profile.

The Comparison Matrix

System	Storage Architecture	Data Type	Execution Specialty
Core Salesforce	Relational Database	Structured Only	Single Record Transactions (OLTP)
Data Warehouse	Columnar DB	Structured/Semi-Structured	Batch Analytics (OLAP)
Data Cloud	Data Lakehouse (Hyperforce)	Structured, Streaming, Unstructured (Vector)	Identity Resolution, Real-time Activation, AI Grounding

The Engineering: DLOs, DMOs, and Data Graphs

Because you cannot simply dump a billion rows into a standard Salesforce Custom Object, Data Cloud uses a distinct three-layer metadata architecture:

• Data Lake Objects (DLOs): The raw ingested data. If you ingest a massive JSON file of web clicks, it lands here exactly as it is.
• Data Model Objects (DMOs): The "Harmonization" layer. You map fields from the messy DLOs into standardized Salesforce Customer 360 models (e.g., mapping "Cust_FName" and "FirstName" both to the standard `First Name` DMO field).
• Data Graphs: The "Pre-computed JSON DataBus." To make retrieving complex relationships lightning fast for prompts or OmniScripts, a Data Graph flattens multiple related DMOs into a single JSON blob.

Data Cloud is Not Just Another Custom Object - This is a Big Misconception!

Architects often think of Data Cloud as just "Big Objects 2.0." It is fundamentally different. Data Cloud does not run on the core Salesforce transactional database. It is a completely separate big-data infrastructure running alongside Core, connected natively via metadata so it looks like it's in Salesforce.

          // How Data Cloud is queried vs Core Salesforce
          // Core SOQL: Synchronous, limited by governor limits.
          List<Account> accs = [SELECT Id FROM Account LIMIT 10];

          // Data Cloud SQL (via API): Asynchronous, hits the Trino engine, can process billions of rows.
          // You do not query Data Cloud with standard SOQL for large aggregations.
          String query = 'SELECT Unified_Individual_Id__c, SUM(Purchase_Amount__c) FROM Sales_Order_DMO GROUP BY Unified_Individual_Id__c';
       

Documentation for Data Cloud Architecture

• Data Cloud Architecture Guide: Understand the separation of storage (AWS S3/Parquet) and compute (Trino).
• Zero-Copy Partner Network Specs: Technical details on how Data Cloud establishes secure, direct network links to Snowflake, Databricks, and Google BigQuery.
• Vector Database in Data Cloud: How unstructured data (PDFs, knowledge articles) is chunked, vectorized, and stored in Data Cloud to power Agentforce RAG (Retrieval-Augmented Generation).