Over the past six decades (since 1964), the mainframe platform has continuously evolved — known over time as the S/370, S/390, and today as IBM Z. It has consistently adapted to industry shifts and market demands, supporting modern programming languages like Java, C, Go, and Python, alongside traditional ones like Cobol and PL/1. Today, it powers essential technologies across hybrid cloud, AI, analytics, DevOps, containers, Linux, and open source, while upholding its fundamental principles of data integrity, reliability, availability, and maintainability.
On June 18, 2025, IBM will introduce the next generation of its mainframe technology: the z17 architecture. What are the key innovations? How does it outperform its predecessors? And how well does it align with today’s IT trends?
One thing is certain: the mainframe is strategically positioning itself for the age of AI.
The z17 can now handle 450 billion inference operations per day, representing a 50% performance boost over the z16. To clarify: in artificial intelligence and machine learning, an inference operation occurs when a trained model processes new input data to make a prediction or decision. While training focuses on learning from data, inference is about applying that learning in real-world scenarios.
Moreover, the z17 is capable of running multiple AI models simultaneously in ensemble configurations, significantly enhancing performance and prediction accuracy. By combining traditional neural networks with large language models (LLMs), the system can tackle complex tasks — for instance, improving fraud detection by layering deep contextual understanding on top of predictive capabilities.
AI is seamlessly integrated across the entire platform stack, enabling real-time, on-site data processing. This results in faster insights and reduced latency — both critical for use cases like fraud prevention and real-time risk analysis.
Designed with adaptability in mind, the z17 can keep pace with the rapidly evolving AI landscape. At launch, the system will support 48 IBM Spyre AI accelerator chips, with the ability to scale to 96 chips within a year. This ensures that organizations can quickly ramp up capacity as newer, larger AI models emerge, especially in domains like fraud detection and risk evaluation.
IBM continues its mission to modernize the mainframe experience with the z17. The goal: to bridge the gap between legacy environments and modern IT requirements. To that end, the z17 offers intuitive tools and features that streamline the user experience. Through familiar interfaces, open standards, and integration with open technologies such as Linux, OpenShift, and Ansible, IBM is lowering barriers for new users while still delivering advanced capabilities to seasoned professionals.
Further enhancements include improved energy efficiency and a next-generation security architecture: With the z17, IBM is introducing IBM Threat Detection for z/OS, a new AI-powered security feature that identifies and mitigates cyber threats early in the attack lifecycle.
Finally, it’s worth noting that the development of the z17 was deeply customer-driven. Through close collaboration with clients via Design Councils and co-creation processes, IBM dedicated approximately 1,600 hours of joint work to shape and refine the z17’s design — reflecting a strong commitment to a customer-centric approach.
