For decades, enterprises have faced the same fundamental challenge: how to connect high-value data in core enterprise systems to modern analytics and AI platforms without compromising the performance, governance, security, or operational integrity on which those systems depend.
At Snowflake Summit 2026, Snowflake reinforced a growing industry shift away from traditional data movement and toward zero-copy interoperability. Enterprise platforms are where some of the most valuable business data lives—and where the pipeline tax has historically been highest. New zero-copy integration capabilities are designed to make data from platforms such as SAP, Salesforce, and Workday available within the Snowflake ecosystem in near real time, without complex ETL pipelines or the loss of critical business context. Deepening partnerships with organizations IBM further extend this model, helping enterprises bring operational and enterprise data together under a more consistent, AI-ready foundation.
That said, of the largest opportunities remains in the systems that have powered enterprise operations for decades. A significant share of enterprise data still resides in mainframe environments—particularly in financial services, healthcare, manufacturing, and the public sector—where these platforms continue to serve as critical systems of record. As organizations pursue more interoperable and AI-ready architectures, extending zero-copy principles to mainframe data has become an increasingly important step toward eliminating data duplication, preserving business context, and unlocking trusted enterprise intelligence at scale.
The Mainframe Data Problem Isn’t Solved (Yet)
Most organizations that have adopted Snowflake have already addressed the relatively easier part of enterprise data integration. They’ve built pipelines from cloud applications, on-prem databases, and SaaS platforms. In many cases, those ingestion patterns are mature, stable, and optimized.
Mainframes, however, remain materially different. These systems, which are often responsible for running mission-critical workloads, also store some of the most valuable transactional data in the enterprise. Accessing and activating that data remains challenging for several reasons:
- Proprietary storage systems, data models, and formats, including IBM Db2 for z/OS and other specialized mainframe data stores.
- Legacy application and processing models, including COBOL workloads and tightly coupled batch architectures
- Architectures optimized first for resilience, throughput, and transactional integrity rather than modern interoperability
- Extraction and replication patterns that can become costly, operationally complex, and difficult to scale over time
Even when organizations succeed in extracting data, it usually passes through intermediate platforms before eventually landing in Snowflake. In some environments, that can mean latency measured in days. If the objective is real-time analytics or AI-driven decision-making, that latency means the data has already lost the lion’s share of what made it valuable in the first place.
The rise of AI and agentic applications, meanwhile, has fundamentally changed the expectations around data freshness. Many modern AI systems cannot rely solely on yesterday’s batch loads. They increasingly depend on real-time or near-real-time operational data, as well as carefully governed and contextualized datasets for reasoning, retrieval, and decision support. For most enterprises, that translates into unified access across a whole host of structured and unstructured sources.
To no one’s surprise (but to everyone’s consternation), this is precisely where the mainframe gap becomes a strategic bottleneck.
The Shift: From Data Movement to Data Access
To support production-grade AI and agentic systems, a meaningful architectural shift is underway in financial services and beyond: from moving and replicating data by default to accessing more of it in place where appropriate.
Snowflake has already helped catalyze this shift with open formats and support for data lake architectures, including Apache Iceberg. This allows external data to be queried without requiring physical ingestion into Snowflake itself. It also aligns with IBM’s broader data and AI strategy, which emphasizes open data ecosystems, hybrid architectures, and federation/virtualization patterns.
IBM’s Role: Virtualizing Mainframe Data for the Cloud Era
IBM has long provided tools for mainframe integration, including technologies such as IBM Data Gate and Db2-related access and replication patterns that make mainframe data more accessible to distributed platforms
These approaches can reduce latency and integration overhead, but in many cases they still rely on some combination of data movement, replication, caching, or duplication.
The newer evolution of the strategy appars in IBM watsonx.data. Rather than positioning it purely as a traditional warehouse, IBM presents watsonx.data as a hybrid data lakehouse and interoperability layer designed to support open, governed data access across environments. Its goal is to:
- Federate or virtualize access to selected enterprise data sources.
- Expose them as open table formats (notably Apache Iceberg) for broader interoperability
- Operate across hybrid and multicloud environments across AWS, Azure, and Google Cloud.
- Preserve governance while enabling interoperability.
In effect, watsonx.data is designed to transform legacy data systems into queryable, open-format datasets without requiring full physical migration.
Turning Open Data into Compute Power with Iceberg and Snowflake
A key enabler in this architecture is Apache Iceberg as an open table format that can decouple shared storage from compute. Where mainframe-derived data can be represented through Iceberg tables and compatible object storage patterns, it becomes more accessible to modern engines such as Snowflake with less reliance on traditional ETL pipelines.
This enables a powerful shift: in the right architecture, data can remain closer to its source, be exposed through modern formats, and be consumed by multiple systems with greater consistency.
On the consumption side, Snowflake is positioned to read and process these open-format datasets directly. Combined with its expanding ecosystem, which includes including capabilities like Snowflake Cortex (for native AI and LLM-driven workloads), Polaris, and Horizon (for governance and cataloging) Snowflake becomes the compute and AI layer sitting on top of open, federated enterprise data.
Streaming the Mainframe: Real-Time Becomes Possible
One of the historic gaps in traditional architectures has been low-latency access to or movement of mainframe data.
With streaming technologies such as Confluent—typically paired with CDC, messaging, or event-enablement patterns—enterprises can move selected mainframe events in near real time, feed operational data into analytics and AI systems, and complement batch and virtualization strategies with live pipelines.
This completes the picture: batch, streaming, and virtualized access all coexisting in a unified architecture. Together, these capabilities can support a more distributed enterprise data architecture—though whether it qualifies as a true data mesh depends on operating model, domain ownership, and governance design. Within that mesh:
- Mainframes remain the system of record.
- IBM technologies provide virtualization and governance.
- Confluent enables real-time event streaming.
- Snowflake delivers scalable analytics and AI compute.
- Open formats like Iceberg act as the interoperability layer.
Crucially, this entire architecture is not about replacing mainframes, but about extending and activating the data they contain to support real-time analytics and agentic workflows.
The Path to the Agentic Enterprise
It goes without saying that the ultimate driver behind this architectural shift is the concurrent shift toward agentic systems and solutions. Modern AI systems are (still) only as effective as the data they can reliably access—and increasingly, that data must be current, well-governed, context-rich, and available across domains.
With Snowflake’s AI and analytics capabilities (including Cortex) combined with IBM’s data virtualization and governance stack, what enterprises are beginning to unlock are use cases like:
- Real-time AI inference directly on operational data.
- Agent-based workflows that span systems of record and systems of insight.
- End-to-end decision automation grounded in trusted, governed enterprise data.
These capabilities help lay the foundation for what many are beginning to call the “agentic enterprise,” where AI does more than analyze information—it participates in workflows across the enterprise
By combining federation and virtualization, open table formats such as Iceberg, streaming platforms such as Confluent, and cloud-scale compute through Snowflake, enterprises are moving toward a model in which less data needs to be moved unnecessarily to deliver value.
Putting Zero-Copy Architecture into Action
The shift from data movement to data access is already reshaping how enterprises think about their most critical systems. What was once viewed primarily as an integration challenge is increasingly becoming a foundational architectural pattern for the AI era—one that brings mainframes, cloud platforms, streaming systems, and open table formats into a more unified and governed ecosystem.
Still, the real test is not the architecture in theory, but the discipline of implementation. Enterprises are now actively working through the practical realities of making this vision real: aligning governance, performance, interoperability, and operating models across environments that were never originally designed to work together.
If you’re exploring how to modernize your data foundation, unlock the value of core systems, or prepare your organization for the next wave of AI innovation, Hakkoda and IBM can help. Reach out to our team of industry-first data experts to discuss your data modernization, AI, and enterprise transformation goals.
