IBM today announced that a mainframe that has been further optimized to run additional artificial intelligence (AI) workloads alongside online transaction processing (OLTP) and analytics applications will become available in June.
Elpida Tzortzatos, an IBM distinguished engineer and IBM z architect, said the IBM z17 includes a faster edition of the Telum processor that IBM developed for offloading AI inference engines from the core mainframe platform.
Dubbed the Telum II, the processor expands the amount of cache made available by 40% to increase the number of AI inference options that could be run on the IBM z16 mainframe that incorporated the previous iteration of the Telum AI accelerator that IBM specifically developed for its mainframe platforms. The iBM z17 can run more than 450 billion inferencing operations in a day with one millisecond response time.
Additionally, IBM in the fourth quarter will make available an IBM Spyre Accelerator for the IBM z17. A system-on-chip (SoC) that IBM has previously previewed, the IBM Spyre Accelerator offloads the running of AI models and associated AI agents from the mainframe.
IBM also committed to developing a new capability to discover and classify sensitive data on the platform using natural language enabled by the Telum II.
At the same time, IBM also previewed z/OS 3.2, an update to the operating system for IBM Z, that will be made available in the third quarter of 2025. In addition to adding support for the latest AI accelerators, it also enables the surfacing of additional AI insights for system management capabilities along with support for NoSQL databases and processing of data across a hybrid cloud computing environment.
IBM also launched IBM Z Operations Unite, which brings together operational data from multiple sources across IBM Z using the OpenTelemetry format being advanced under the auspices of the Cloud Native Computing Foundation (CNCF). IBM Z Operations Unite will reduce the amount of time needed to detect anomalies by a factor of nine, according to IBM. The watsonx Assistant for Z will be integrated with Z Operations Unite to provide AI chat-based incident detection and resolution for the first time using live systems data.
IBM is also making an edition of the Vault platform for managing secrets that was originally developed by HashiCorp — a company IBM acquired earlier this year — on the z17.
Finally, IBM is making available an updated edition of the IBM Storage DS8000 array that is optimized for the IBM z17.
Collectively, these capabilities will enable organizations to run lower-latency AI applications that surface more accurate outputs, said Tzortzatos. Additionally, the mainframe will consume less energy, improve security and enable better governance over the data used to train AI models in an on-premises IT environment, she added.
It’s not clear how many AI models are being built and deployed on IBM mainframes but given the amount of data running on these platforms, it’s all but certain that just about every application running on the most venerable of platforms in the enterprise will soon be infused with AI.
