IBM has announced the release of its latest mainframe hardware, designed to accelerate the adoption of artificial intelligence (AI) technologies.

On Monday, the company unveiled the IBM z17, the newest iteration of its mainframe computer hardware. Powered by the IBM Telum II processor, this fully encrypted mainframe is capable of supporting over 250 AI use cases, including AI agents and generative AI, according to IBM.

Despite being perceived as outdated, mainframes remain a crucial component of modern computing, with 71% of Fortune 500 companies currently utilizing them, according to a report. The mainframe market was valued at an estimated $5.3 billion in 2024, as reported by Market Research Future, a consulting firm.

The z17 boasts a significant improvement in processing power, capable of handling 450 billion inference operations per day, which represents a 50% increase over its predecessor, the IBM z16, released in 2022. The z17 is designed to seamlessly integrate with other hardware, software, and open-source tools. The IBM z16 was powered by the company’s original Telum processor.

Tina Tarquinio, Vice President of Product Management and Design for IBM Z, revealed to TechCrunch that the development of the z17 mainframe upgrade began five years ago, prior to the current AI revolution sparked by the release of OpenAI’s ChatGPT in November 2022.

IBM invested over 2,000 research hours in gathering feedback from more than 100 customers during the development of the z17, according to Tarquinio. She noted that it is interesting to see how the feedback they received aligned with the current market trends.

“The introduction of our AI accelerator has coincided with significant changes in the industry regarding AI, particularly in the latter half of 2022,” Tarquinio told TechCrunch. “It’s been exciting to witness. The possibilities are truly endless when it comes to what AI can help us achieve.”

The z17 is designed to adapt to the evolving AI market, Tarquinio explained. Upon release, the mainframe will support 48 IBM Spyre AI accelerator chips, with plans to increase this number to 96 within the next 12 months.

“We intentionally built in headroom and AI agility to accommodate new models as they emerge,” Tarquinio said. “As new models are introduced, we want to ensure that we have the necessary headroom for larger models that may require more local memory to interact with each other. We’ve built this in because we know that the approach will continue to change, with new models being developed and old ones being retired.”

Tarquinio highlighted the energy efficiency of the z17 as one of its key features, joking that it was like choosing a favorite child. The z17 is reportedly more energy-efficient than its predecessor and competing systems, Tarquinio claimed.

“We’re increasing AI acceleration by seven and a half times on-chip, while using five and a half times less energy than other accelerators or platforms in the industry,” Tarquinio explained.

The IBM z17 mainframes are scheduled to become generally available on June 8.