For years, enterprise data architecture has relied on collecting, transforming, and replicating data across multiple systems to serve different analytical and operational needs. This approach has delivered value, but it has also introduced unwanted side effects, over time: unnecessarily duplicated datasets, rising storage and processing costs, and a data governance landscape that becomes increasingly complex as the number of platforms grows. Essentially, modern pipelines attempt to normalize the erroneous idea that data must be constantly duplicated just to make it usable. However, every copy introduces latency, cost, and inconsistency. In recent years, to overcome these side effects, Zero-Copy Architecture (ZCA) has been gaining significant attention.
It is increasingly viewed as a natural response to the growing need for more efficient, governed, and unified access to data across complex enterprise environments. ZCA aims to break the assumption that “moving data is inevitable.”
However, ZCA has actually been in existence for more than two decades, and the most complete implementation of this architecture is the Denodo Platform, which is built on a logical approach to data management. Let’s break down why that is.
What Zero-Copy Architecture Truly Means
ZCA is best understood as a logical approach to managing and delivering data, rather than a specific product or deployment pattern. Its core idea is simple: organizations should be able to provide consistent, governed, and timely access to data without relying on unnecessary physical replication.
In ZCA, data consumers—whether analytical tools, operational systems, or AI workloads—interact with a shared, authoritative representation of the data, even when that data originates from diverse systems and storage technologies.
This requires several capabilities working together:
- Efficient access to data where it resides, so that physical relocation becomes an option rather than the default.
- Optimization techniques that minimize data movement while still delivering strong performance.
- A semantic layer that provides unified meaning and structure across heterogeneous sources.
- Consistent governance and security controls that apply uniformly, regardless of where the data is stored.
- Flexible delivery channels such as SQL, APIs, and AI-ready interfaces, so that the same governed data can support different types of consumers.
The long-held aspiration of consolidating all enterprise data into a single system has gradually given way to a more pragmatic view: modern architectures must work with data wherever it naturally lives. A ZCA embraces this reality by reducing friction, accelerating access, and avoiding the operational overhead traditionally associated with repeated data copies.
Why Are Open Table Formats Alone not a Real Zero-Copy Solution?
Iceberg and Delta solve important storage-layer issues such as ACID transactions, schema evolution, or time travel, but they still operate under a fundamental constraint: they assume all important data lives inside the lake.
That’s rarely true in the real world, where critical systems remain in:
- Operational databases
- SaaS platforms
- Mainframes
- API-driven services
- Legacy systems that cannot be migrated
- Document stores
- Cloud warehouses outside your main cloud, and other data sources
Zero-Copy cannot require a massive migration just to start working. If your architecture begins with “First put everything into object storage,” then it’s not Zero-Copy—it’s Copy-First.
Recent articles argue that Zero-Copy can be achieved simply by combining open table formats (Iceberg, Delta, Hudi…), shared catalogs like Unity or Glue, and federated engines such as Trino, Snowflake External Tables, or Databricks Federation. But this describes only a very narrow scenario: reducing duplication inside the lake. It does not solve Zero-Copy across SaaS systems, operational databases, mainframes, API-driven services, multicloud environments, or external partners.
Open table formats standardize files, not business meaning, cross-system governance, semantic layers, lineage, or enterprise-wide unification of disparate sources. They are necessary building blocks, but insufficient for Zero-Copy at the enterprise level.
Denodo: The Real Zero-Copy (or Logical-First) Architecture
Denodo starts from a completely different premise: don’t move data unless there is an unavoidable reason to do so. It provides four capabilities no modern lakehouse stack can match in combination:
- Native access to any data source: Denodo connects to relational databases, NoSQL stores, SaaS APIs, cloud warehouses, object storage, streams, and even legacy systems—all without relocating the data. This immediately eliminates the need to copy data into a lake as a prerequisite for analysis.
- Distributed optimization and deep pushdown: Denodo doesn’t simply federate queries; it optimizes them. It rewrites queries, pushes filters and joins to remote systems, prunes partitions, leverages indexes, and minimizes data movement to an absolute minimum. This turns Zero-Copy into a performance-viable strategy instead of a theoretical aspiration.
- A complete semantic layer: Formats like Iceberg or Glue Catalog describe files —not business meaning. Denodo builds reusable business views, hierarchies, canonical models and domain-level abstractions, giving teams a shared understanding of the data without creating new physical datasets. This is what enables multiple teams to consume the same asset without divergence.
- Unified governance, security, and lineage: When data is replicated across platforms, governance becomes fragmented. Denodo centralizes RBAC and ABAC security through row-level security, column-level security, masking policies, permission models… including auditing and lineage—all in one place. This prevents Zero-Copy from turning into Zero-Control.
Zero-copy beyond analytics: APIs, applications, and GenAI
ZCA is most valuable when it can support a wide range of use cases, not just analytical querying. Modern organizations increasingly rely on governed, consistent data to power applications, microservices, APIs, operational dashboards, automation flows, and emerging AI capabilities.
Denodo contributes to this broader landscape by enabling the same logical data models to be accessed through multiple delivery interfaces—including REST, GraphQL, OData, JDBC, and ODBC. In addition, its integration with the Denodo AI-SDK, and MCP support, enables large language models (LLMs) and other AI tools to interact with governed data directly, without needing to generate or maintain additional physical copies.
This approach extends ZCA principles into operational and AI-oriented scenarios, enabling consistent, governed data to be used across the enterprise without expanding the footprint of data replication.
How Denodo goes further than the Lakehouse approach
Lakehouse-driven ZCA is fundamentally storage-centric. It assumes that if you standardize storage and metadata, add a query engine, and layer in some data governance, you get ZCA “for free.”
In practice, this solves only a small portion of the problem. It provides ZCA inside the lakehouse, but nowhere else.
If you need Zero-Copy between SaaS systems, multiple clouds, operational, and analytical systems, Zero-Copy governance for APIs, apps or AI, or Zero-Copy semantics across domains, the lakehouse stack falls short. It becomes a partial solution that still requires massive ingestion projects and synchronization pipelines.
Denodo, by contrast, starts from the point of maximum heterogeneity. It unifies data wherever the data lives, making the lakehouse simply another source instead of the mandatory center of the universe. While lakehouse stacks try to approximate Zero-Copy by eliminating duplication inside object storage, Denodo implements Zero-Copy at the enterprise level, across all systems, cloud systems, tools, and domains. That’s the difference between a storage architecture and a logical data management architecture.
Zero-Copy Isn’t a Trend
As cloud costs rise, AI requires governed access to broader datasets, and multi-cloud becomes the default reality, so replicating data everywhere becomes financially and technically unsustainable. Zero-Copy is not just a “nice-to-have”: It’s the only long-term architecture that scales.
And among all platforms claiming to enable it, only Denodo delivers the full picture: from query pushdown to governance, semantic modeling, API exposure, GenAI integration—without forcing you into a single storage pattern.
The future of data architecture will belong to approaches that prioritize logic over location, governance over duplication, and intelligence over brute-force movement of data. Zero-Copy embraces this evolution—and positions organizations to grow, innovate, and adapt without being constrained by where their data happens to live.
Others preach “Zero-Copy” but require copying everything into their storage layer first, turning the idea of a Zero-Copy Architecture into a contradiction—a “Copy-First” approach in disguise. Logical-First is the real Zero-Copy: data access and governance without moving data, unless there is a real, justified need.
- Introducing Logical-First Architecture, the Real “Zero-Copy” Architecture - February 11, 2026
- Data Mesh: Challenges and Solutions - October 19, 2023