Martyn Rhisiart Jones

Segovia 20th December 2025

The Data Warehouse Is Dead. Long Live the Data Warehouse.

In 1992, Bill Inmon coined the term “data warehouse” and laid out four sacred rules: subject-oriented, integrated, non-volatile, time-variant. It was a blueprint for a fortress of truth, expensive, on-premises, batch-processed, and utterly indispensable. Fast-forward three decades. The fortress has been replaced by something that resembles a hyperscale cloud platform. This platform can run your AI models and your CEO’s dashboard simultaneously. Welcome to data warehousing in 2025.

The transformation has been tectonic. What used to be racks of hardware humming in a basement are now fully managed, serverless platforms. They scale compute and storage independently. These platforms charge you only for what you actually use. They let you ingest petabytes of streaming data without breaking a sweat. Snowflake, Google BigQuery, Amazon Redshift, Microsoft Fabric. These are the new names in town. They are rewriting the rules. Meanwhile, they quietly preserve the old ones.

Data volumes have exploded. Half of the data is semi-structured JSON or Parquet. The other half is increasingly unstructured and screaming for vector embeddings. Business leaders no longer wait until tomorrow morning for yesterday’s numbers; they want insights before the coffee is cold. And cloud computing has turned fixed-cost infrastructure into a utility you can dial up or down like a thermostat.

The result is a modern data warehouse. It looks very different from Inmon’s original vision. However, it still obeys Inmon’s core commandments.

It is still subject-oriented. Data is organized around business domains such as customers, sales, and inventory. This organization uses star schemas, snowflake schemas, or semantic layers. These layers are built with tools like dbt and Looker.

It is still integrated. ETL/ELT pipelines, data catalogues, and governance frameworks ensure that disparate sources become a single source of truth. Examples include Snowflake Horizon and BigQuery Data Catalogue.

It is still time-variant. Timestamped tables and Type 2 slowly changing dimensions are examples. Features like Snowflake’s TIME TRAVEL or BigQuery’s SYSTEM_TIME let you ask “what did we know on March 15?”

And it is still non-volatile, at least mostly. ACID transactions now allow controlled updates and deletes. (Hello, GDPR.) However, the analytical core remains append-only or change-tracked. It preserves history as sacredly as ever. The adaptations are what make the difference. Real-time streaming (Kafka connectors, Snowflake Streams, BigQuery Streaming) replaces overnight batch jobs. In-database ML and vector search power AI applications without moving data. Self-service semantic layers let business users explore data without having to beg IT for a new report. Storage and compute are decoupled. You can store exabytes cheaply. You can spin up thousands of concurrent queries without rewriting the budget.

So, does the modern data warehouse still follow Inmon’s principles? Yes. Unequivocally. The spirit lives on, even if the body has been rebuilt from the ground up.

The catch is that this evolution has made the platform choice more consequential than ever. If you choose incorrectly, you’ll be locked into a single cloud provider. You may be saddled with unpredictable costs. You could also be stuck with a system that can’t keep up with tomorrow’s AI workloads. Pick the right one, and you get a platform that delivers enterprise-grade performance. It provides ironclad governance and predictable pricing. It remains accessible to everyone from data engineers to the marketing team.

In 2025/2026, the data warehouse isn’t dead. It’s just been reborn as something faster, cheaper, and far more potent than Inmon might have imagined. The old rules still hold. The new ones are just a lot more fun.

Many thanks for reading.

The 100 most recent articles on goodstrat

The Last 100 Good Strat Articles in One Place