IBM launched its IBM z17 mainframe that includes AI tools to process inference workloads, its Telum II processor and Spyre Accelerator as well as AI agents from its watsonx platform.
While many folks think of mainframes as creaky old systems designed to process transactions without downtime, the z systems are big business for IBM and a strong upgrade cycle. In addition, IBM has reinvented the mainframe a few times for cloud workloads and now AI.
IBM's bet is that the z17, which will be available June 18, can take on new workloads beyond transaction processing. IBM said the system can score 100% of their transactions in real time and process 50% more AI inference operations per day than the z16.
According to IBM, the z17 can handle more than 250 AI use cases including loan risk, managing chatbot services, supporting medical image analysis and curbing retail shrink.
For IBM, the z17 is also a showcase for its integrated stack of R&D, software, AI and hardware.
Components include:
- Telium II processor, which has a second-gen on-chip AI accelerator with a 1 millisecond response time.
- IBM Spyre Accelerator that will be available in the fourth quarter of 2025 via PCIe card. Spyre will bring genAI tools to the mainframe and run assistants leveraging data on the system.
- Spyre will enable z17 to run a host of IBM Granite models natively.
- A series of AI agents and assistants from IBM watsonx including watsonx Code Assistant for Z and Assistant for Z. Watsonx Assistant for Z will be integrated into Z Operations Unite for AI-driven incident detection and resolution.
- Z Operations Unite, available in May, combines logs from IBM Z in OpenTelemetry format to streamline operations and detect anomalies.
- IBM will also include HashiCorp tools to standardize secrets management.
- z17 will also include data security tools via Telium to identify and protect data with natural language.
In addition to the AI focus, IBM z17 will feature z/OS 3.2 that will be released in the third quarter. The operating system will support AI capabilities, modern data access methods, NoSQL databases and hybrid cloud data processing.
Constellation Research's take
R "Ray" Wang, CEO of Constellation Research, said:
"While many may have written off the mainframe, there are three reasons this release is significant:
- Y2Q is closer than we think and the Z is quantum secure.
- Cost per MIPS/kwh gives the Z very efficient performance.
- AI is one area where a return to on-premises is a real value prop.
When speaking with existing customers who are using mainframe, they are mostly looking to increase their investment."
Constellation Research analyst Holger Mueller said:
"The typical adage of hardware platforms going to die does not apply to the IBM mainframe. Z has had at least 7 lives and is now back in the AI era. And its core value proposition of bringing data and processing power into one place - is even more attractive in the AI era compared to long passed client server era. Kudos goes to IBM for innovating and delivering value on System Z for enterprises over decades (APIs, Java, hybrid cloud and more come to mind as highlights of the last 20 years)."
