Cover Image for IBM launches a new mainframe system designed for the era of artificial intelligence.
Tue Apr 08 2025

IBM launches a new mainframe system designed for the era of artificial intelligence.

The IBM z17 mainframe has the capability to perform 450 billion artificial intelligence inferences per day, representing a significant advancement compared to its previous model.

IBM has announced the launch of the latest version of its mainframe hardware, which includes updates designed to facilitate the adoption of artificial intelligence. The new model, named IBM z17, is equipped with an IBM Telum II processor and targets over 250 AI use cases, including AI agents and content generation. Although mainframes may seem like a technology from the past, 71% of Fortune 500 companies still use them.

According to market estimates, the mainframe sector will reach a value of approximately $5.3 billion by 2024. The z17 system is capable of executing 450 billion inference operations per day, representing a 50% increase over its predecessor, the IBM z16, which was released in 2022 and used the company's original Tellum processor.

This new mainframe has been designed to fully integrate with other hardware, software, and open-source tools. Tina Tarquinio, Vice President of Product Management and Design for IBM Z, mentioned that this mainframe upgrade has been in development for five years, before the current surge in AI that began with the launch of ChatGPT in November 2022. Tarquinio noted that IBM dedicated over 2,000 hours of research to gather feedback from more than 100 clients during the creation process of the z17 and highlighted how the needs reflected in that feedback align with current market trends.

IBM Z's VP emphasized that the z17 is ready to adapt to the evolving AI market and will be equipped with 48 IBM Spyre AI accelerator chips at launch, with plans to double that number in the next 12 months. "We are designing with room to maneuver," Tarquinio stated, pointing out that provisions have been made to accommodate larger models that may require additional local memory for internal communication.

Another highlighted aspect of the new hardware is its energy efficiency, which compares favorably not only with its predecessor but also with competing products. Tarquinio explained that the z17 increases AI acceleration by seven and a half times while using five times less energy compared to other types of accelerators in the industry.

The z17 mainframes will be generally available starting June 8th.