Do Mainframes Have A Role In The AI Era?

Forbes

10-04-2025

AI-Generated, AI-Enabled Mainframe
Francis Sideco
A couple of years after generative AI first entered mainstream consciousness, every industry segment is attempting to leverage it to improve efficiency and offer new products and services. While training will continue to evolve and drive innovation, inferencing will drive value creation through advanced AI capabilities such as chain of thought reasoning, multi-modal functionality and multi-model support combining different types of generative AI models with predictive AI. Most of the conversations on delivering these inferencing solutions gravitate around AI data centers, edge infrastructure, and/or on-device processing. This begs the question: Does the mainframe have a role in the AI Era?
What Is A Mainframe?
While servers are designed more for supporting general-purpose applications and multiple clients or functions like website hosting and email servers, mainframes are designed for high-volume, mission-critical tasks such as financial transaction processing and are often used in heavily regulated industries. As such, mainframes require a higher degree of capacity, reliability and security enabled with advanced virtualization, disaster recovery, backwards compatibility and built-in redundancy. Additionally, workloads are typically handled by a centralized mainframe system whereas a distributed architecture to spread workloads over many systems is commonly used in a server architecture.
With the capacity, reliability and security that mainframes provide along with their ubiquity in supporting high-volume, high-value transactions and data processing, the answer to whether mainframes have a role in the AI Era is an unequivocal 'Yes!' IBM, the leader in mainframe solutions with 70% of all global financial transactions going through their mainframes, is a prime example.
A Mainframe For The AI Era
Telum II By The Numbers
Francis Sideco
IBM recently announced the latest in its Z family of mainframes, the Z17, with the goal of addressing the needs of the AI Era while still delivering on the rigorous expectations associated with mainframes. According to IBM, this Z17 generation is powered by its 5nm 5.5GHz Telum II CPU, which, compared to the previous generation, delivers an 11% increase in single-thread performance, up to 20% capacity expansion and up to 64 TB of memory, while also doing so with up to 27% power reduction. Additionally, Telum II has an enhanced on-board AI accelerator capable of predictive and some generative AI workloads. For the generative AI workloads that require more acceleration, the Z17 can also be upgraded with the new Spyre Accelerator PCIe card.
Spyre Accelerator By The Numbers
Francis Sideco
Based on IBMs reported performance numbers, the Z17 provides 7.5x more AI throughput than the Z16 generation delivering up to 450 billion AI inferences with 1ms response times per day.
What AI Workloads Need A Mainframe?
Due to their heavy use in financial transactions and mission-critical data processing, mainframes are most effective when using a combination of predictive and generative AI models. For example, because of the high-volume, central processing, and multi-model capabilities, mainframes can effectively and efficiently analyze patterns from the transactions and data passing through the system and infer conclusions that can be used in advanced fraud detection and anti-money laundering applications for improved accuracy and fewer false positives. Mainframes also combine these capabilities with mission-critical business data to help enable business, code and operations assistants to increase productivity and reduce the time needed for skills training, and autonomous agentic AI applications like automated trading and healthcare applications. Other areas where AI-enabled mainframes are being used include, but are not limited tom loan risk mitigation, insurance claims fraud detection and prevention, payments fraud, geospatial analysis, climate change impact, loan risk mitigation, cybersecurity and sentiment analysis.
These are just a small subset of applications where predictive, generative and even agentic AI leverage the mainframe for business outcomes that would otherwise be inefficient or not available because of the data and/or security requirements on standard server configurations especially in heavily regulated industries in which mainframes are typically deployed.
The Future Of Mainframes In The AI Era
According to IBM, there are already more than 250 client-identified AI use cases on the Z mainframe platform and growing. But it's not all about the hardware. IBM leverages its other AI assets like watsonx, Granite, InstructLab and even their consulting services, across IBMs solutions, including the Z platform, positioning the company as a strong partner for the age of enterprise AI. Competitors such as Dell, Fujitsu, and Unisys are also looking to leverage AI for mainframe workloads. Next generation mainframe development typically takes 5-7 years and if they're anything like IBM, it is safe to say that the next few generations of AI-enabled mainframes are already in the works. Not only are mainframes surviving in the AI Era, they are thriving.