Every year, the world of Data and AI seems to reinvent itself. Few industries embody the principles of the VUCA world—volatility, uncertainty, complexity, and ambiguity—quite as strongly as this one.
New terms, new hypes, new methods – and yet, one old question remains:
Do I still need to model data?
Will Data Warehouses still exist in the future, or will AI do the modeling for us?
The goal of this article is to clarify the most important terms and show what role data modeling actually plays in today’s modern data and AI landscape.
The Many Terms: Data Lakes, Lakehouse, DWH & More
How can we make sense of all the buzzwords—Data Lake, Data Lakehouse, classic Data Warehouse (DWH), or Enterprise DWH (EDWH)?
And when data is modeled, should we follow the bottom-up approach by Kimball or the top-down approach by Inmon?
Is there an overarching concept that unites them all?
Or can we skip modeling altogether thanks to modern ideas like Data Products and Data Assets?
Self-Organizing Data Architectures? A Myth.
If you listen to today’s vendor marketing presentations or read current AI-related articles, you might get the impression that data architectures now organize themselves – and that data modeling has become obsolete.
But that is, and remains, an illusion.
Is the Data Warehouse Still Relevant?
The discussion around Kimball, Inmon, and Data Vault often appears as a methodological dispute. Yet it’s worth focusing on what they have in common:
All approaches aim to create a structured core model that organizes a company’s data objects, and on top of it, a reporting or interface layer optimized for analytics.
Without this layered structure, data can hardly be meaningfully related—and it’s precisely here that value is created.
Data Vault emphasizes automation (DWA – Data Warehouse Automation) and, with Data Vault 2.0, seeks to unify the classical models within an integrated methodology.
The Data Lakehouse, meanwhile, combines the principles of the DWH and the Data Lake—an interesting twist, considering that Data Lakes were once seen as the replacement for the DWH.
Today, we know better: the DWH hasn’t disappeared—it has become part of a more evolved architecture.
Data Assets, Data Products & Data Mesh
Concepts like Data Products or Data Mesh operate on a higher organizational and management level—they don’t replace data modeling or a DWH.
“Raw” and “Combined” Data Products instead represent an iterative approach: start simple, and add layers and models as they become necessary.
In practice, these ideas are less about replacing classical modeling and more about prioritization. After all, almost every organization already faces a modeling backlog.
Data Mesh expands on these ideas both organizationally and technically—it aligns data ownership with concrete products.
And if a data product requires modeled data—and it usually does—those models must be provided within the corresponding Data Asset.
A Raw Product alone rarely generates significant business value.
Hybrid AI and Data Modeling
Even in the era of large language models (LLMs), data modeling remains indispensable. Because of their probabilistic nature, LLMs have limited applicability in enterprise contexts: hallucinations and forgetfulness make additional structures like Vector RAG or Graph RAG necessary—databases that store and leverage business context.
But even within these hybrid AI systems, traditional modeling still takes place.
Experts must still understand how data and business processes relate and organize them intentionally to generate value.
The effort remains—the LLM is not the architect but rather the tool operating on top of modeled data.
Conclusion
All of the concepts discussed—Data Lakehouse, Data Mesh, Data Vault, Data Products, and others—contribute to making data architectures more modern, efficient, and flexible.
But none of them eliminate the need for data modeling.
The Data Warehouse is far from obsolete.
It continues to live on—in new forms—as part of modern architectures, as the foundation for Data Products, and as the backbone of AI systems.
Artificial Intelligence can support modeling, suggest entities from natural language, recognize patterns, or document models.
But the actual task—structuring data, defining relationships and keys, and anticipating future requirements—remains a matter of human expertise.
Data modeling is not a relic of the past.
It is—and will remain—the foundation of every value-generating data architecture.
