Built for the age of generative and agentic AI.

At an event held at IBM’s One Madison Avenue auditorium in New York City with over 100 clients in attendance and more than 15,000 viewing online, IBM announced IBM z17 to the world. 

Extending a ten-year run of Z growth, the new z17 is built from silicon to software for AI workloads and hybrid cloud, helping enterprises unlock the full potential of their data, accelerate ROI, and enable business innovation and growth.

Fun facts about the Z mainframe

• Did you know the letter Z in the Z mainframe family stands for “zero downtime”?
• IBM Z is known for its 99.999999% availability (315 milliseconds/year). That’s one hour of downtime every 11,000 years!

A modern mainframe for AI and mission critical workloads

Powered by the new IBM Telum II processor with advanced on-chip inferencing, IBM z17 helps bring AI to their mission-critical data and transactions. z17’s latest advances allow 50% more AI in-transaction inferencing than z16, without sacrificing reliability, security or sustainability. 

Clients can extend and scale this capability in z17 with the new Spyre accelerator cards, available in 4Q 2025. Together, Telum II and Spyre support a broader set of AI models and use cases, promising significant accuracy and performance gains. 

Both are built for high-speed inferencing at the source of data. The IBM Telum II processor offers on-chip AI acceleration and is optimized for extremely low latency within the system. The Spyre Accelerator is geared to handle larger, more complex AI use cases, including generative AI and multimodel methods. Multiple Spyre Accelerators can be connected into the I/O subsystem of IBM Z via PCIe-attach. Combining these two technologies can result in a substantial increase in the amount of available acceleration.

Beyond fraud detection, some other typical AI use cases on z17 include mitigating loan risks with improved AI prediction accuracy, analyzing medical images with better precision and defending data against cyber threats, among many others.

z17 by the numbers

Enhancing user productivity with assistants and agents  

z17 accelerates mainframe application modernization initiatives with generative AI, supporting automation tools like AI assistants and agents, including IBM watsonx Code Assistant for Z and IBM watsonx Assistant for Z. In this way, z17 bolsters the skills and efficiency of developers and IT operations with capabilities for application discovery and analysis, code explanation, automated code refactoring, COBOL to Java transformation, and more. In addition, watsonx Assistant for Z will be integrated with the new IBM Z Operations Unite, which brings operational data from multiple sources across IBM Z to streamline operations and accelerate the time to detect anomalies. 

Putting security and cyber defense at the core 

Another attribute of the IBM z17 that clients value highly is its strong security and resiliency. New capabilities include integration with HashiCorp, an IBM Company, to help standardize secrets management across hybrid cloud. The new IBM Vault Self-Managed for IBM Z provides a centralized way to control access to credentials, API keys, certificates, encryption keys and other secrets, reducing security risks, such as unauthorized access, hardcoded secrets and secrets sprawl.  
 
z17 also continues to help clients prepare for a post-quantum world with technology built in, including NIST-standardized quantum-resistant cryptographic algorithms, helping ensure robust data protection against future threats. 
 
Also in plan is a new capability to discover and classify sensitive data on the platform. This would tap into Telum II for natural language processing allowing sensitive data to be identified and protected before being used in the AI data pipeline.  

Why it matters

z17 is a cornerstone of enterprise IT and integral to the backbone of the global economy for decades. A resilient technology foundation is essential for organizations to better prepare for the future and z17 is engineered to enable organizations to run AI models along with the most demanding enterprise workloads, without compromising on throughput or inferencing quailty.