Google Cloud launches AI tools for mainframe migration

Google Cloud has introduced an artificial intelligence-based approach to mainframe migration and modernisation aimed at large organisations with complex legacy estates.

It is presenting the offering as an alternative to either keeping ageing mainframe systems in place or attempting a single large-scale migration. Many enterprise projects go beyond simple code conversion because older applications are often tied to proprietary databases, file formats, transaction monitors and operational tools.

Google Cloud has grouped the offering into four areas: assessment, application modernisation, pre-production testing and data migration. The portfolio combines its Gemini models with existing products for mainframe environments.

Assessment first

The process begins with the Mainframe Assessment Tool, which reverse-engineers legacy applications. It maps dependencies between applications and data stores, extracts business rules from older code, generates technical documentation and identifies application boundaries across business functions.

Google Cloud argues this stage is necessary because large mainframe estates often contain obscure relationships between programs, databases and workflows that are not visible through source code conversion alone. In practice, those can include links to DB2 databases, VSAM files, IMS structures and tightly integrated transaction processing systems.

By documenting those relationships, the tool is intended to give engineering teams a clearer picture of how existing systems work before any migration begins. That information can then feed into automated workflows used to plan or execute modernisation work.

Two routes

Customers can choose between two broad migration patterns depending on the workload. One focuses on rewriting or reimagining applications when companies want to change business logic or redesign customer-facing systems. The other is a deterministic, like-for-like modernisation path intended to preserve existing behaviour while changing the underlying application structure.

The first route is aimed at businesses that want to move beyond legacy design and create a new cloud-native application. Under that model, the assessment tool extracts business rules, while specialised agents help produce target specifications, architecture designs and implementation plans.

The second route keeps the application's external behaviour unchanged. It is designed for workloads where organisations want to reduce technical debt or move away from the cost and operational constraints of a mainframe without altering outputs or interfaces.

To illustrate the split, Google Cloud pointed to financial services as an example of where mixed strategies may be used across the same estate. Stable back-office batch jobs could move through a like-for-like process, while systems viewed as competitive differentiators might be rebuilt in a different form.

Reducing risk

A key part of the package is Dual Run, a product designed to compare the output of old and new systems before cutover. It runs production workloads in parallel across the mainframe and a Google Cloud environment, then checks whether transactions, messages and data changes match.

That side-by-side testing is meant to address one of the biggest obstacles in mainframe projects: proving that a new system behaves in the same way as a long-established one under real production conditions. The process continues until logic and data equivalence is achieved.

Google Cloud added that some global enterprises are already using Dual Run to support migration projects and meet regulatory approval requirements in tightly controlled sectors. It did not identify those customers.

Data transfer

The fourth element is the Mainframe Connector, which is designed to copy data from mainframe systems into cloud services including BigQuery, Spanner, Cloud SQL and Cloud Storage. The connector handles codebase and data type conversion and can be integrated into existing extraction, transformation and loading processes.

That gives customers a way to move selected data incrementally rather than through a single transfer. It also reflects a broader industry shift, as companies look to use long-held operational data for analytics and new digital services without waiting for an entire core system to be replaced.

Mainframe modernisation remains a significant market for cloud providers and consulting firms because many banks, insurers, governments and large industrial groups still run critical workloads on decades-old systems. Those environments often support high-volume transaction processing, but they can be difficult to change because of specialist languages, bespoke interfaces and a shortage of experienced staff.

Google Cloud's approach suggests the next phase of competition in that market will centre less on simple code translation and more on proving that AI tools can interpret business logic, preserve operational reliability and lower the risk of failure in large, interdependent estates.

Google Cloud said a single transaction scenario in some mainframe environments can involve millions of lines of code.