What is a data warehouse?
Lightbridge Cloud defines a data warehouse as a central repository that integrates data from many source systems into one structured store, optimized for analytics, reporting, and business intelligence rather than day-to-day transactions. Its classic properties are subject-oriented, integrated, time-variant, and non-volatile, which is what makes it the durable system for decision-making across an organization.
A data warehouse is a central, analytics-optimized repository that integrates data from many source systems.
A data warehouse is a database built for analysis rather than transactions. It consolidates data from operational systems such as CRM, ERP, finance, and marketing into one consistent store, cleansed and modeled so that reports and dashboards run against a single trusted source. This analytical purpose, online analytical processing, or OLAP, is what separates a data warehouse from the transactional (OLTP) systems that run day-to-day operations.
The classic definition, from Bill Inmon, gives a data warehouse four properties: it is subject-oriented, organized around business subjects like customers and revenue; integrated, reconciling the formats and keys of every source; time-variant, retaining history so trends can be measured; and non-volatile, stable once loaded so analysts query consistent data. Those four properties still describe what a warehouse is for, whether it runs on-premises or in the cloud.
Lightbridge Cloud advises on data warehousing as an independent, vendor-neutral practice. This guide explains the concept, the surrounding architecture, and the platform landscape. For the commercial practice, including how Lightbridge scopes and delivers data and integration work, see the integration services page.
A data warehouse, a data lake, and a data lakehouse solve related but distinct problems.
A data warehouse stores structured, curated data and applies a schema when data is written, schema-on-write, which makes it query-ready and governed for business intelligence. A data lake takes the opposite approach: it stores raw data of any format on commodity object storage and applies structure only when the data is read, schema-on-read. The lake is flexible and scales for large, varied data, including inputs for data science; the warehouse is the trusted, modeled store for reporting.
A data lakehouse is the hybrid that has emerged to bridge the two. It adds warehouse-style management, ACID transactions, schema enforcement, governance, and time travel, over commodity lake storage using open table formats such as Delta Lake and Apache Iceberg. Databricks popularized the lakehouse around Delta Lake, and platforms including Snowflake now support Iceberg tables, so the historical line between lake and warehouse is blurring. Many organizations still run a lake and a warehouse together; a lakehouse consolidates both onto one architecture when that fits the workload.
A data warehouse is populated through ETL or ELT pipelines from its source systems.
Data reaches a warehouse through pipelines that extract, transform, and load it. The two patterns differ in the order of those steps. Traditional ETL, extract, transform, load, transforms data in a separate processing tier before it lands, so the warehouse receives data already shaped and cleansed. Modern cloud ELT, extract, load, transform, loads raw data into the warehouse first and then transforms it in place using the warehouse's own compute.
Cloud data warehouses made ELT practical, because elastic compute can transform large volumes efficiently once the data has landed, and keeping a raw copy speeds initial loading and preserves the source for reprocessing. ETL still suits cases where data must be governed and cleansed before it lands, or where the target has limited transformation capacity. Both remain valid; most cloud designs today favor ELT. These pipelines are integration work, covered in the iPaaS guide, and where the source is a system of record such as an ERP, that system is owned by the Lightbridge ERP practice.
Data warehouse modeling uses dimensional star schemas and normalized enterprise designs.
How a data warehouse is modeled determines how well it performs and how easily people can query it. The dimensional approach, associated with Ralph Kimball, organizes data into fact tables that hold measurable events, such as sales, and dimension tables that describe the context, such as customer, product, and date. When dimensions surround a central fact table, the layout is called a star schema. Star schemas are intuitive for business users and efficient for the aggregation and filtering that reporting requires.
The alternative, from Bill Inmon, is a normalized enterprise data warehouse: a highly structured, integrated core that acts as the single source of truth, from which dimensional data marts are then derived for consumption. The two approaches are complementary rather than opposed. Many organizations combine a normalized core with dimensional star schemas at the reporting layer, choosing the balance that fits their governance needs and the way their analysts work.
A cloud data warehouse relies on MPP, columnar storage, and separated storage and compute.
Cloud data warehouses earn their scale from three architectural ideas. Massively parallel processing (MPP) splits a query across many nodes that work simultaneously, so scanning large volumes stays fast. Columnar storage keeps data by column instead of by row, which is far more efficient for the aggregations and selective reads that analytics performs, and it compresses well. Together they let a warehouse answer heavy analytical queries over large history without the row-by-row overhead of a transactional database.
The defining shift in the cloud era is the separation of storage and compute. An organization can scale query power up or down independently of how much data it stores, run multiple isolated compute clusters against the same data, and pay for compute only while queries run. This elastic model is what lets a cloud data warehouse absorb spiky, unpredictable analytical workloads. Lightbridge Cloud helps design and size this architecture against the real volumes and query patterns in play.
A data warehouse powers business intelligence, historical analysis, and data science.
A data warehouse earns its place when an organization needs one trusted, historical view of data spread across many systems. These are the patterns that recur most often. The unifying theme is a single governed store that every team reports from, replacing conflicting extracts and per-system reporting with consistent, auditable answers.
Business intelligence and reporting
The primary use of a data warehouse is powering dashboards, scorecards, and reports across finance, sales, and operations. A single governed store means every team reports off the same numbers, ending the reconciliation disputes that arise when each function queries its own operational system.
Consolidating source systems
A data warehouse brings together CRM, ERP, marketing, support, and web data into one model. Pulling from a system of record such as an ERP is a common source; ERP advisory, selection, and implementation are owned by the Lightbridge ERP practice, while Lightbridge Cloud builds the data layer around it.
Historical and trend analysis
Because a data warehouse retains history, it answers questions operational systems cannot: how revenue trended over three years, how churn moved by cohort, how a metric compares to the same period last year. Time-variant storage is what makes this analysis possible.
Self-service analytics
A curated, well-modeled warehouse lets analysts and business users explore data through BI tools without writing pipelines or touching production systems. Governance and a clear semantic model keep that self-service accurate rather than a source of conflicting one-off extracts.
Feeding data science and AI
A data warehouse provides the clean, integrated, historical data that machine learning and forecasting models depend on. Many teams use the warehouse, or a paired data lake, as the feature store and training source that grounds analytics and AI in consistent, governed data.
Regulatory and financial reporting
A non-volatile, auditable warehouse supports the repeatable, point-in-time reporting that finance and compliance require. Because loads are stable and history is retained, a report can be reproduced exactly, which matters when numbers must be defended to auditors or regulators.
Where a warehouse draws from an ERP or other system of record, that source system, including ERP advisory, selection, and implementation, belongs to the Lightbridge ERP practice. See lightbridgeerp.com for that side. Lightbridge Cloud builds the data and integration layer that consolidates those systems into a warehouse.
The defining characteristics of a data warehouse set it apart from an operational database.
What distinguishes a data warehouse from the transactional systems it draws on is a set of deliberate design choices. These are the characteristics an organization should expect from a warehouse, and the ones to weigh when comparing platforms and architectures against its own workloads.
Analytics-optimized, not transactional
A data warehouse serves online analytical processing (OLAP): aggregations, trends, and reporting across large history. This is the opposite of the online transactional processing (OLTP) that runs an application, where the priority is fast single-row reads and writes. Separating the two keeps analytics from slowing the systems that run the business.
Integrated from many sources
A data warehouse consolidates data from CRM, finance, operations, and other systems into one consistent model. Integration reconciles the different formats, keys, and definitions each source uses, so a single query can join across the whole organization instead of pulling from isolated silos.
Subject-oriented and curated
Data is organized around business subjects such as customers, orders, or revenue rather than around the source application. The data is cleansed and conformed on the way in, which is what makes a data warehouse a trusted source for reporting instead of a copy of raw operational tables.
Time-variant history
A data warehouse retains history so it can answer how a metric looked last quarter or last year, not just its current value. Time-variance is what supports trend analysis, period-over-period comparison, and the point-in-time reporting that operational systems, which overwrite in place, cannot provide.
Non-volatile store
Once loaded, data in a warehouse is stable: it is read and appended rather than continuously updated in place. Non-volatility gives analysts a consistent target to query, so two reports run against the same load return the same numbers rather than shifting mid-analysis.
Modeled for query performance
A data warehouse structures data deliberately, typically with dimensional models and conformed schemas, so common analytical queries run efficiently. Modern cloud warehouses add columnar storage and massively parallel processing to scan large volumes fast, which is what keeps reporting responsive at scale.
Leading cloud data warehouse platforms include Snowflake, BigQuery, Redshift, and Databricks.
The cloud data warehouse market includes several established platforms, each strong in a different dimension. Snowflake is known for near-zero administration and consistent operation across AWS, Azure, and GCP. Google BigQuery is fully serverless, allocating compute per query with no cluster to size. Amazon Redshift is the AWS-native choice for organizations already on that cloud. Databricks is a lakehouse platform strong in data engineering and machine learning, and Microsoft consolidates analytics through Azure Synapse Analytics and the newer Microsoft Fabric.
The right platform depends on the data volume and workloads, the existing cloud footprint, the in-house skills available, and total cost across compute and storage. There is no single best data warehouse, only the best fit for a given environment. The line between warehouse and lakehouse continues to blur as platforms adopt open table formats such as Apache Iceberg, which is one more reason the choice should follow workload rather than label. For the integration pipelines that feed any of these platforms, see the iPaaS guide.
Lightbridge Cloud is an independent, vendor-neutral data advisor. It is not a partner tier or reseller of any data warehouse platform, so its recommendation follows fit rather than commission. Platform and product names are trademarks of their respective owners; Lightbridge Cloud is independent and is not affiliated with, endorsed by, or a partner tier of any of them.
Lightbridge Cloud advises on and delivers data warehouse work as a vendor-neutral practice.
Lightbridge Cloud helps organizations decide whether a data warehouse, a data lake, or a lakehouse is the right model, select a platform on fit rather than partner incentives, and then design the architecture, data model, and the ETL or ELT pipelines that populate it. Because Lightbridge does not carry a partner-tier product quota for any data warehouse vendor, the recommendation follows what an environment actually needs rather than what a reseller is paid to push.
This page explains the concept. The engagement model, scope, and delivery approach live on the integration services page, the pipelines that feed a warehouse are covered in the iPaaS guide, and ERP source systems are owned by the Lightbridge ERP practice. Those are the right next steps for an organization moving from understanding a data warehouse to building one.
Data warehouse: frequently asked questions
- What is a data warehouse in simple terms?
- A data warehouse is a central repository that pulls data from many different systems and stores it in one place, organized for analysis and reporting. In simple terms, it is where an organization keeps a clean, combined, historical copy of its data so people can run reports, build dashboards, and answer business questions. It is built for analytics rather than for running applications, so heavy reporting queries do not slow down the systems that handle day-to-day transactions. Its classic properties are that it is subject-oriented, integrated, time-variant, and non-volatile. Lightbridge Cloud advises on data warehousing as a vendor-neutral practice, recommending the platform and design that fit an environment rather than a single product.
- What is the difference between a data warehouse and a data lake?
- A data warehouse stores structured, cleansed, and modeled data, applying a schema when data is written (schema-on-write), which makes it curated and query-ready for business intelligence. A data lake stores raw data of any format, structured, semi-structured, or unstructured, on commodity object storage and applies structure only when the data is read (schema-on-read). The warehouse is the trusted, governed store for reporting; the lake is the flexible, high-volume store for large and varied data, including inputs for data science. Many organizations run both, landing raw data in a lake and serving curated data from a warehouse. A data lakehouse aims to combine the two on one architecture.
- What is a data lakehouse?
- A data lakehouse is a hybrid architecture that adds data warehouse management features over commodity data lake storage. It layers ACID transactions, schema enforcement, governance, and time travel onto open table formats such as Delta Lake and Apache Iceberg, so raw object storage can behave like a database. The goal is to run warehouse-style analytics and data science on one copy of data, rather than maintaining a separate lake and warehouse. Databricks popularized the term around Delta Lake, and platforms including Snowflake now support Iceberg tables, blurring the historical line between lake and warehouse. A lakehouse suits teams that want a single platform for both curated BI and large-scale, open-format data.
- What is the difference between ETL and ELT?
- ETL and ELT describe the order of the same three steps: extract, transform, and load. In traditional ETL, data is extracted from sources, transformed in a separate processing tier, and then loaded into the data warehouse already shaped for use. In modern cloud ELT, raw data is extracted and loaded into the warehouse first, then transformed in place using the warehouse's own compute. Cloud warehouses made ELT practical because elastic, separated compute can transform large volumes efficiently once the data has landed. ELT keeps a raw copy and speeds initial loading; ETL applies governance and cleansing before data lands. Most cloud data warehouse designs today favor ELT, though both patterns remain valid depending on the workload.
- What is a star schema?
- A star schema is a dimensional modeling design, associated with Ralph Kimball, that organizes a data warehouse into fact tables and dimension tables. A fact table holds the measurable events, such as sales transactions with quantities and amounts, and it references surrounding dimension tables that describe the context, such as customer, product, date, and location. The layout resembles a star, with the fact table at the center and dimensions radiating out. Star schemas are widely used because they are intuitive for business users and efficient for the aggregation and filtering that reporting requires. This dimensional approach is complementary to Bill Inmon's normalized enterprise data warehouse model, and many organizations combine a normalized core with dimensional marts for consumption.
- What is a cloud data warehouse architecture?
- A cloud data warehouse is built on three architectural ideas that give it scale and elasticity. Massively parallel processing (MPP) splits a query across many nodes that work in parallel, so large scans complete quickly. Columnar storage keeps data by column rather than by row, which is far more efficient for the aggregations and selective reads that analytics performs. Separation of storage and compute lets an organization scale query power up or down independently of how much data it stores, and pay for compute only while queries run. Together these let a cloud data warehouse handle large, spiky analytical workloads without the fixed capacity of an on-premises appliance. Lightbridge Cloud helps design and size this architecture for a given workload.
- Which are the leading cloud data warehouse platforms?
- Platforms frequently cited as leading cloud data warehouses include Snowflake, Google BigQuery, Amazon Redshift, Databricks, and Microsoft, through Azure Synapse Analytics and the newer Microsoft Fabric. Each has a distinct profile: Snowflake for near-zero administration and multi-cloud consistency, BigQuery for a fully serverless model, Redshift for AWS-native workloads, and Databricks for a lakehouse strong in data engineering and machine learning. Microsoft Fabric consolidates analytics on the Microsoft stack. The right choice depends on the data volume and workloads, existing cloud footprint, in-house skills, and total cost across compute and storage. Lightbridge Cloud is an independent, vendor-neutral data advisor and is not a partner tier or reseller of any of these platforms, so its recommendation follows fit rather than commission.
- How does Lightbridge Cloud help with a data warehouse?
- Lightbridge Cloud is an independent, vendor-neutral cloud and data advisor. It helps organizations decide whether a warehouse, a data lake, or a lakehouse fits their needs, select a platform on fit rather than partner incentives, and design the architecture, data model, and ETL or ELT pipelines that populate it. Because Lightbridge does not carry a partner-tier product quota for any data warehouse vendor, its recommendation follows what an environment actually needs. Warehouses are usually fed by integration pipelines and often draw from a system of record; the integration side is covered in the iPaaS guide and the integration services page, and ERP source systems are owned by the Lightbridge ERP practice.
From understanding a data warehouse to a working analytics platform.
When the question shifts from what a data warehouse is to which platform fits and how to build it, Lightbridge Cloud runs a vendor-neutral assessment and designs the architecture and pipelines that turn scattered data into trusted analytics.