Data architecture is the framework that defines how data is collected, stored, transformed, and consumed within an organization. It provides the technological foundation needed by data systems, establishing rules, policies, standards, and models that govern which data is collected and how it’s stored, arranged, integrated, and used in data systems. A well-designed data architecture ensures data is accessible, secure, high-quality, and aligned with business goals.
Why centralized architectures fall short
The dominant argument for data centralization is operational efficiency: bringing all data to one place, controlling access, simplifying discovery, and enforcing governance at scale. However, in real-world organizations, especially those that are large or diversified, data naturally becomes dispersed.
Example
A global retailer with operations spanning North America, Europe, and Asia may use different point-of-sale systems, comply with regional privacy regimes, and operate data centers both on-premises and in the cloud. A mandate to centralize all transactions, customer data, and inventory into a single U.S.-based data warehouse triggers massive complexity:
- Data sovereignty rules in the EU prohibit the transfer of personal data outside the European Union in certain contexts.
- Business units resist forced migration due to the risk of service interruptions and duplicated integration effort.
- Some business applications generate unstructured or streaming data that is unsuitable for the warehouse’s tabular schema.
The practical result: Centralization projects face delays and budget overruns while data silos persist, now with added friction.
Centralization can delay, not accelerate, AI adoption
AI systems, particularly generative AI and agentic AI architectures, rely on rapid, high-quality, and context-rich access to data, regardless of its location. The traditional approach—copying and standardizing all enterprise data into a single centralized repository—creates significant bottlenecks, from slow ingestion and transformation cycles to data staleness and project delays.
Example
A financial services company wants to deploy an internal chatbot that provides instant insight into both customer profile data (stored on-premises) and transaction history (in a SaaS CRM). A centralized data model requires extracting data from both systems, harmonizing formats, and continuously synchronizing updates—a process that can take months to fully operationalize and is prone to lag.
When the chatbot is asked about a customer’s recent loan payments, it may return outdated answers if the latest transactions have not yet been loaded. Meanwhile, a federated approach—where the AI queries each system directly—can deliver timely and accurate responses, thereby accelerating innovation and time-to-value.
Centralization increases the risk of shadow IT and data silos
Rigid centralization often encourages teams to circumvent official channels when business needs are urgent. This results in a chaotic approach to IT, where departments copy data to spreadsheets, build unauthorized databases, or provision unsanctioned SaaS tools to get their work done.
Example
A healthcare provider with a centralized patient data warehouse imposes a slow, approval-heavy ETL process for every update. The marketing department, under pressure to launch a new outreach campaign, extracts patient contact info and campaign data into their own cloud workbook—outside of established governance. Data quality, accuracy, and security can no longer be assured, and departmental silos are worsening rather than improving.
Centralized architectures struggle with data diversity and velocity
The era of big data, now augmented by AI, demands support for a range of data types (structured, semi-structured, and unstructured) and sources (on-premises, multi-cloud, and SaaS). Centralized systems, particularly older warehouses, struggle with rapidly evolving data formats or high-velocity streaming data.
Example
A logistics firm wants to build an AI-driven route optimization model that ingests real-time vehicle telemetry (IoT), weather APIs (semi-structured data), and legacy inventory data (relational database). A central data warehouse often enforces rigid schemas and slow batch ingestion, resulting in stale insights and difficulty handling unstructured or streaming data at the required scale. As a result, the AI model operates on partial or outdated input, reducing its effectiveness.
Centralization increases cost and operational overhead
Migrating and maintaining vast amounts of diverse data in a single platform is expensive. Storage, compute, maintenance, and ETL costs can balloon, especially as data volume outpaces engineering bandwidth. Additionally, proprietary centralized storage tends to cause vendor lock-in, limiting flexibility.
Example
A manufacturing conglomerate begins migrating legacy ERP, CRM, and sensor data to a cloud data warehouse. The cost of continuous ETL, data transformation, and storage—especially for rarely accessed data—mounts rapidly. Licensing and egress fees make the organization dependent on a single vendor, undermining its long-term bargaining power and technical agility.
Centralization can complicate regulatory compliance
Many organizations operate under regulations that require defined data residency, auditability, and fine-grained access controls—for example, GDPR (Europe), HIPAA (health), PCI-DSS (payments), and the EU Data Act. Centralized approaches can struggle to enforce such policies across diverse geographies or data domains.
Example
A multinational insurance company must respond to a “right to be forgotten” request from a European customer. If the customer’s data exists in both a centralized U.S. data warehouse and replicated data lakes or data lakehouses in the EU, verifying and deleting every reference is a complex, time-consuming, and error-prone process. Keeping sensitive data in decentralized, well-governed local domains while providing federated access enables compliance-by-design without requiring global data movements.
Centralization hinders business agility
As business teams increasingly demand self-service analytics and AI, the centralization model—with its reliance on centralized engineering or data stewardship—often fails to deliver data at the required speed or in a business-relevant context. The need to wait for centralized approval or pipeline changes slows innovation.
Example
A consumer goods company’s marketing team wants to experiment with new customer segmentation for a campaign. Under a centralized model, they must submit a ticket for IT to update the warehouse ETL logic, wait for the next sprint, and validate the output before starting their analysis. Valuable insights may be missed due to the inherent delays of centralization.
Overcoming centralization: Open lakehouse and hybrid approaches
Modern data architecture is shifting toward hybrid models that blend selective centralization with federated and decentralized access. Open data lakehouse architectures—such as those built on Trino and Apache Iceberg—illustrate the new paradigm.
What is an open data lakehouse?
An open lakehouse is a modern hybrid architecture that combines the best features of data lakes and data warehouses. It provides storage for both structured and unstructured data (similar to a data lake) while also offering data management features, ACID transactions, and performance optimizations (akin to a data warehouse). Built on open standards and formats, it enables universal data access across on-premises, multi-cloud, and SaaS data sources without requiring full centralization, making it well-suited for AI workloads and meeting regulatory compliance requirements.
This model offers:
- Universal data access: Direct querying across on-premises, multi-cloud, and SaaS data sources, without requiring full centralization.
- Governance without compromise: Role- and attribute-based access controls, consistent auditing, and data lineage across decentralized domains.
- Choice and flexibility: Avoidance of lock-in through open formats and engines, with the ability to move only critical data to central storage when it makes sense.
- Real-time and AI readiness: Support for high-velocity data ingestion and a mix of structured/unstructured data, ideal for analytics and AI.
- Business-driven collaboration: Easy packaging and sharing of high-quality data products, breaking down barriers between teams while maintaining compliance.
Consider a pharmaceutical firm with regulated research data on-premises (to comply with intellectual property and HIPAA requirements), marketing analytics in a cloud data lakehouse, and finance data in a SaaS platform. By deploying a federated Icehouse platform:
- Scientists can run AI-powered drug discovery models on local, governed datasets without uploading sensitive IP to the cloud.
- Marketing analysts query up-to-date sales and campaign results across clouds, joining them with field data collected via mobile apps.
- Finance teams can run consolidated risk models that draw from both the SaaS platform and on-premises sources, governed by a unified policy, even as datasets remain physically distributed.
This architecture supports business agility, AI enablement, and regulatory compliance—something challenging with traditional centralization.
Conclusion
Centralized data architectures thrived in earlier eras when data was simpler, less distributed, and analytics requirements were static. Today, with AI driving new demands for access, velocity, and context—and with regulatory, organizational, and technical complexity at an all-time high—their limitations are apparent.
Forcing all enterprise data into a single repository can lead to delays, silos, compliance challenges, mounting costs, and a loss of business agility. The future belongs to open, hybrid, federated architectures—particularly open data lakehouses—which blend the strengths of centralization (when needed) with the flexibility and pragmatism data teams require.
Organizations that recognize and adapt to these limits—opting for universal data access, open formats, and governance across a hybrid backbone—will be best positioned for the AI-driven future, driving innovation, compliance, and competitive advantage.
