IBM Spyre Accelerator, generally available for IBM z17 on 28 October 2025, is set to revolutionize how organizations apply AI in their enterprises. This powerful AI accelerator is designed to bring generative AI to the IBM z17, enabling enterprises to scale up large language model (LLM) AI capabilities within the trusted envelope of IBM Z environments.
With Spyre Accelerator, clients will have the ability to use LLMs with their internal operations, integrating functions in a natural language interface such as watsonx Assistant for Z. As clients assess and build for the future, agentic AI will become a critical tool for efficiency and modernization of process flows across the data center. Similarly for business, Spyre Accelerators are engineered to help the future development of higher accuracy AI techniques, allowing for building out generative asset creation, vendor or customer data ingestion and interpretation for outreach, cross-offers, or even risk assessments, real-time.
The benefits of combining predictive and generative AI techniques to create an enterprise-level portfolio across the full spectrum of encoder, decoder and encoder-decoder AI models.
One client was very clear about the intersection of core trusted environments with enterprise-level objectives. “On-prem with Spyre Accelerator for IBM Z is important to us because this work will be related to production code,” said the Infrastructure lead at a Large European bank. “We don’t want the production workloads leaving our hands, so it is our favorite option.”
AI-driven tools can help clients quickly predict next actions needed to support a busy transaction day or incorporate recommended agents and APIs to get optimized output from a new application, accelerating the time to value for the business. According to a study from the Institute for Business Value and Oxford Economics in 2024, 61% of executives find generative AI crucial for application modernization efforts on mainframes.
As part of continuous deliveries, key software products like IBM watsonx Assistant for Z, AI Toolkit for IBM Z and IBM LinuxONE, and Machine Learning for IBM z/OS will use Spyre accelerators for full on-prem deployment. To accomplish on-prem LLM support, each Spyre Accelerator contains 32 AI-optimized processing cores to support large language models directly on the mainframe, enabling enterprises to run secured generative AI use cases on-premises. This infrastructure is designed to be scalable as PCIe-attached adapters, manageable as part of the environments already handling core data. This marks a significant milestone, allowing businesses to leverage AI at scale while keeping data on IBM Z.