Data Mesh 2.0

Introduction

“If data is the new oil, why do so many enterprises still spill it before it reaches the refinery?”

Over the past two decades, enterprises have been on a relentless pursuit to harness data. This journey began with centralized data warehouses, evolved into data lakes, and eventually progressed to lakehouses—each promising to eliminate silos, reduce duplication, and unlock enterprise-wide intelligence.

However, despite significant engineering effort and investment, many organisations still struggle to make data discoverable, usable, and trustworthy in line with modern business demands.

The root cause? Centralized bottlenecks.

When all data flows into the hands of a single team—whether a central data engineering group or an analytics Center of Excellence (COE)—queues grow, priorities clash, and domain experts are disconnected from shaping the data they rely on. This creates a paradox in which massive infrastructure investments coexist with slow and constrained decision-making.

In 2019, Zhamak Dehghani introduced Data Mesh as a radical rethinking of enterprise data architecture:

• Domain-oriented ownership: Enable domain teams to own and manage their data.
• Data as a product: Treat datasets as products with clear ownership and quality standards.
• Self-serve data platforms: Empower teams to work independently without relying on central gatekeepers.
• Federated governance: Establish standards without centralized control.

While Data Mesh 1.0 addressed key organisational bottlenecks, it revealed limitations—particularly when viewed through the lens of AI-era requirements:

• How can real-time compliance be ensured when dozens of domains manage their own pipelines?
• How can data quality be automated without relying on large teams of data stewards?
• How can AI copilots and generative models be supported with high-quality, context-rich data at scale?

This is where Data Mesh 2.0 comes into focus. It retains the decentralized philosophy while introducing AI-driven governance, automated metadata discovery, federated machine learning, and composable data product frameworks. Rather than a replacement, it represents a refinement—designed for scale and suited to multi-cloud environments, IoT ecosystems, and real-time analytics.

India stands at a unique inflection point. Digital Public Infrastructure (DPI) initiatives such as Aadhaar, UPI, and ONDC already embody mesh-like decentralization, making the transition to Data Mesh 2.0 a natural progression. In banking, HDFC Bank employs decentralized credit risk analytics aligned with mesh principles. In healthcare, Apollo Hospitals is experimenting with domain-owned patient care datasets. In manufacturing, Tata Steel leverages IoT-driven data domains for predictive maintenance.

As global enterprises such as JPMorgan Chase, Roche, and Netflix continue to refine their mesh architectures, Indian organizations have the opportunity to learn, adapt, and leap directly to Data Mesh 2.0 best practices—avoiding the “half-mesh” pitfalls experienced by early adopters.

The question is no longer “Should we decentralize?”
It is now: “Are we ready to decentralize intelligently?”