Mainframes have been accompanied by predictions of their disappearance for decades, but a seemingly small new feature in Linux 7.2 tells a different story. The s390 architecture has incorporated support for Rust within the kernel, enabling one of Linux’s most recent initiatives to develop low-level components with enhanced memory safety guarantees to also be available on the environment where current IBM Z systems operate.
The key points of mainframe modernization in 30 seconds
- Linux 7.2 introduces Rust support for s390, the architecture used by current IBM Z systems.
- The name s390 originates from an IBM mainframe family from the 1990s, but Linux retains it for the current architecture.
- An Accenture study cited by AWS shows that 90% of organizations surveyed are pursuing cloud modernization efforts, yet only 19.5% of workloads have been migrated.
- AWS, IBM, and Google are already using AI to analyze, document, and transform legacy applications.
- The trend indicates progressive modernization where understanding existing software is as important as moving it.
The name can be confusing. IBM System/390 appeared in 1990, and this family of machines was later replaced by zSeries and the current IBM Z. However, Linux continues to refer to this architecture as s390.
That’s why the arrival of Rust is more significant than it appears.
The code sent to the kernel configures s390 as a 64-bit architecture capable of utilizing Linux’s Rust infrastructure and includes specific components necessary for features such as notice and error management, static branches, and binding generation. This support was developed by Jan Polensky from IBM and was integrated into the updates for Linux 7.2.
It doesn’t mean Linux on IBM Z will be rewritten in Rust, nor that the vast existing C code will disappear. The key point is that new kernel components written in Rust can now start to take this architecture into account.
Rust has been advancing within Linux for years precisely because it helps prevent certain categories of errors related to memory management by design. Its extension to s390 demonstrates that mainframes remain part of the kernel’s technological evolution, not just a legacy platform to be maintained until retirement.
90% want to modernize, but only 19.5% of workloads have been migrated
The paradox of mainframes becomes evident when looking at what happens within large enterprises.
An Accenture survey from 2025, conducted among 1,000 executives and later summarized in an analysis with AWS published in July 2026, found that 90% of mainframe-using organizations are developing cloud modernization initiatives.
However, only 19.5% of their workloads have completed the migration process.
The gap between these percentages helps clarify the issue.
Deciding that an application needs modernization is relatively simple. Figuring out what it actually does after 30 or 40 years of modifications can be much more complicated.
A bank, insurer, or government agency may have COBOL applications running for decades, accumulating changes along the way. Some of their logic may be documented, but other parts reside solely in the code itself, batch jobs, JCL, procedures, databases, and dependencies on other systems.
Even a seemingly straightforward application may contain business rules introduced many years ago, authored by someone who no longer works at the organization.
Rewriting software without understanding these dependencies introduces a clear risk: the new program may be technically correct but behave differently in specific edge cases that the original system was designed to handle over decades.
This is where artificial intelligence is finding one of its less glamorous but potentially most practical business uses.
AWS, IBM, and Google bring AI to legacy code
Leading technology providers already have tools specifically aimed at this problem.
AWS Transform for Mainframe became generally available in May 2025. The service uses AI to assist at various stages of modernization, from analysis and initial planning to documentation generation and refactoring. AWS has also added tools to analyze dependencies and query generated documentation via AI from the code.
IBM watsonx Code Assistant for Z adopts a similar approach within the IBM ecosystem. It can analyze applications, explain code in natural language, document business logic, and participate in refactoring processes. Its features include generating Java services from COBOL and JUnit tests to verify semantic equivalence between the two.
Google Cloud Mainframe Modernization utilizes Gemini models and specialized agents for reverse engineering, assessment, and application transformation. Its Mainframe Assessment Tool can extract business rules, document applications, identify dependencies, and generate test cases.
Google also offers Dual Run, a tool that allows running production traffic on both the original mainframe and the modernized application to compare results before fully switching over.
A common denominator among these approaches is: before transforming the code, you must understand it.
AI can accelerate one of the most labor-intensive phases in such projects—traversing large codebases, generating documentation, reconstructing dependencies, explaining programs written in less common languages, and extracting business rules for review.
This is different from expecting a model to process millions of lines of COBOL and automatically produce a production-ready microservices architecture.
Modernization may mean keeping part of the mainframe
For years, modernization has been associated with migration: removing applications from the mainframe to run on distributed servers or in the cloud.
But current strategies offer more options.
An organization can maintain certain processes on IBM Z, expose functionalities via APIs, decouple components, move specific data, shift some applications, and decommission others. It can also run the old and new systems in parallel temporarily until there’s enough evidence that both behave identically.
This progressive approach reduces one of the greatest risks of a large migration: attempting to replace a system with decades of business logic all at once.
The very fact that Rust is now part of s390 adds another chapter to this story. While part of the industry develops AI tools to understand and transform legacy applications, the Linux community continues to incorporate new technologies into the architecture used by modern mainframes.
These movements are not contradictory.
Organizations can modernize applications without immediately abandoning the platform they run on. They can also modernize how they develop software for the mainframe itself.
Perhaps that’s why the disappearance of the mainframe has been predicted for so long without actually happening. Many of these systems contain applications that are difficult to replace, but that doesn’t mean everything around them must stay untouched.
The question is becoming less binary: choosing between keeping the mainframe or shutting it down. Modernization can also mean discovering what should remain, what can be decoupled, what’s worth transforming, and what can finally be retired. AI doesn’t eliminate this work but can make the crucial step beforehand—understanding exactly what the software that an organization has been running for decades actually does—much more manageable.
Frequently Asked Questions
Does Linux 7.2 support Rust for mainframes?
Yes. Among the changes in Linux 7.2 is the enablement of Rust support for the s390 architecture, used by Linux on current IBM Z systems.
Why does Linux still call IBM Z architecture s390?
The name comes from IBM System/390, but Linux retains the s390 name for the architecture that evolved into the current IBM Z systems.
Can AI automatically convert COBOL into modern applications?
There are tools capable of analyzing, explaining, and transforming code, but enterprise modernization also requires validation of dependencies, business rules, behavior, and functional equivalence. IBM and Google, for example, incorporate test generation and validation mechanisms within their tools.
Are companies rapidly abandoning their mainframes?
Available data suggests a far more gradual process. The Accenture survey cited by AWS indicates that 90% of organizations are developing cloud modernization initiatives, but only 19.5% of workloads have been fully migrated.
