The mainframe computer turned 60 last year. For decades, these room-sized machines have been the reliable but unglamorous workhorses powering large organisations, especially banks and government offices. But far from being retired, the platform is staging a comeback as artificial intelligence (AI) redefines the value of real-time and secure computing.
Case in point: 78% of senior business and IT leaders globally believe mainframes remain essential to their operations, according to the 2025 State of Mainframe Modernization report by IT infrastructure services provider Kyndryl. At the same time, 88% are deploying or planning to deploy generative AI tools and applications in their mainframe environments, and they expect this will drive billions in cost savings and increased revenues.
“AI is extending the relevance of the mainframe in areas where latency, privacy and always-on operations are non-negotiable. Customers increasingly use AI to inform decisions directly where transactions happen, such as spotting fraud or approving credit in real time, without moving sensitive data off-platform. This is keeping the mainframe central to business-critical operations,” Petra Goude, Kyndryl’s president of Strategic Markets, tells The Edge Singapore.
Combining mainframe data with AI can also help organisations unlock new revenue streams. “The largest monetisation opportunities come from real-time decisioning at the point of interaction, rather than back-office analytics,” says Goude. Airlines, for example, can use AI-driven bundles and pricing powered by mainframe passenger data to improve yields and boost ancillary revenue.
Going hybrid
Despite their benefits, mainframes are no longer standalone platforms. Organisations are modernising mission-critical operations through hybrid IT strategies that blend mainframes with cloud computing. The Kyndryl study shows companies are moving about 28% of workloads off mainframes to balance the platform’s performance with the flexibility of cloud services.
See also: Singapore Academy of Law brings AI to 35-year-old LawNet platform
“Mainframe and cloud are complementary. It comes down to gaining the most value from all your platforms,” says Goude. Mainframes excel where AI must process data with extreme throughput, confidentiality and auditability, making it ideal for supporting banking or insurance decisions that must be made in milliseconds using highly sensitive information. Meanwhile, less sensitive or experimental work is shifting to the cloud, where AI services can be built, tested and refined before being deployed back to mainframe environments.
In practice, most enterprises are choosing a hybrid approach: innovate quickly in the cloud, then bring insights and models into the mainframe for secure, real-time execution.Petra Goude, president of Strategic Markets, Kyndyl
The hybrid model is being reinforced by regulatory pressure. As governments impose stricter rules on where and how data is stored and processed, mainframes are emerging as what Goude calls a “sovereign core” within hybrid IT architectures. They allow organisations to keep control over residency, access and usage, while still connecting to private cloud capabilities for innovation.
See also: AI is eating software, and the world
“Enterprises that will thrive will be those that blend both worlds: productising AI on their core systems, including mainframes, while harnessing hyperscaler [or cloud] ecosystems for speed. AI embedded in the mainframe will remain vital for digital sovereignty and secure, real-time decision-making, [while] hybrid cloud environments will be essential for experimentation and scale,” she adds.
Cost factors
The hybrid approach could also help keep IT costs in check. Gartner forecasts public cloud spending will grow by over 20% this year, but many companies here struggle with visibility and predictability in their bills, notes Andrew Lim, Kyndryl’s managing director for Asean and Korea.
“The answer isn’t to walk away from cloud, but to design hybrid architectures with financial operations (FinOps) practices to manage spend. At the same time, mainframes provide predictable cost structures and energy savings, which are critical in markets where power prices are volatile,” he says.
Goude advises organisations to look beyond short-term savings and weigh several interconnected factors when deciding where to run AI workloads. They include:
- Workload fit — Matching the platform to the applications that will be running.
- Data costs — Keeping AI close to the mainframe avoids expensive transfers to the cloud.
- Energy savings — Consolidating workloads on new-generation mainframes can cut power use by up to 75%.
- Risk costs — Stronger security and compliance reduce the financial impact of breaches or fines.
Ultimately, organisations should focus on resilience, or the ability to sustain AI growth without runaway expenses while ensuring sovereignty and compliance. “It is essential to embark on a holistic modernisation journey. We help our customers modernise their mainframe and other legacy systems by taking a ‘right workload on the right platform’ approach…to support their business outcomes,” she says.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Raising the bar on security
Security is another defining factor in the mainframe’s resurgence. Since mainframes encrypt data by default, isolate workloads, and now include quantum-safe security features, they are well-suited to protect AI-driven processes.
“Embedding AI into mainframe environments raises the bar on security. Enterprises are managing against the tampering or misuse of AI models by carefully designing, implementing, and operating AI solutions, while placing strong emphasis on data governance, integrity, and clear AI policies," says Goude.
She continues: "Anyone embarking on this journey must think about cybersecurity as part of the foundation when developing and deploying AI models. Consequently, deep expertise that combines mainframe, cybersecurity, and AI skills becomes essential.”
Lim reinforces that point by highlighting Zero Trust as the next step.
We see Zero Trust as a strategic approach to protecting critical assets by integrating all available capabilities in a coordinated and adaptive way.Andrew Lim, managing director for Asean and Korea, Kyndryl
Kyndryl has been helping businesses adopt Zero Trust in practice by integrating technologies, such as secure access service edge (SASE) and AI, into security frameworks tailored to their IT environments and operational needs. This includes the Kyndryl Agentic AI framework, which is designed to help businesses in Singapore and across the region deploy and scale AI with Zero Trust principles at the core.
Lim also stresses the need for unified security operations as enterprises modernise mission-critical systems from mainframes to the cloud. Doing so will provide visibility, automation and resilience across the entire IT estate.
To address this, Kyndryl is working with LifeTech to implement a next-generation security operations centre platform. It is built on Kyndryl’s Security Intelligence and Automation framework, and is designed to strengthen detection, response and operational efficiency.
Bridging the skills gap
Making mainframe data accessible for AI is challenged by both data and skills. This is because mainframe data often sits in complex, older formats like COBOL that need to be exposed through application programming interfaces (APIs) for AI to use. Moreover, 46% of people entering the workforce do not have mainframe skills, while 39% of experienced staff are retiring and taking their skills with them, according to Kyndryl’s study.
The good news is that generative AI can help address this mainframe skills gap. Generative AI can translate COBOL code into modern programming languages, bringing greater consistency and reliability to what has long been a complex, labour-intensive task. It can also generate technical documentation for legacy applications, which is often scarce but essential for enabling new functions and integrating with cloud platforms. In addition, generative AI can suggest improvements to code quality and system architecture, helping programmers save time and control costs while maintaining, refactoring or re-engineering COBOL-based systems.
“AI or generative AI will continue to be a strong component to support any system modernisation. By combining human expertise with AI, organisations can accelerate mainframe modernisation and make better use of their data [for AI],” says Goude.
While AI is the shiny new tool capturing boardroom attention, the old workhorse of enterprise IT could prove more valuable. The mainframe’s resurgence signals that companies must modernise their IT foundation to harness AI at scale without sacrificing resilience.
Source: Kyndryl's 2025 State of Mainframe Modernization report
