TLDR: IBM has announced the general availability of its new Spyre Accelerator, a dedicated AI inference engine designed to deliver low-latency performance for generative and agentic AI workloads on IBM Z, LinuxONE, and Power systems. This innovation aims to enable enterprises to integrate AI with existing mission-critical data securely and efficiently on-premises, addressing the growing computational demands of modern AI.
IBM is set to revolutionize enterprise artificial intelligence with the introduction of its Spyre Accelerator, a purpose-built AI inference engine designed for low-latency processing of generative and agentic AI workloads. This significant development, announced with general availability slated for late October and early December 2025, targets IBM Z, LinuxONE, and Power systems, emphasizing responsiveness, security, and uptime for mission-critical AI applications.
The Spyre Accelerator is engineered to allow enterprises to integrate advanced AI capabilities directly with their existing systems of record, ensuring sensitive data remains on-platform without compromising performance. This on-premises solution addresses the critical need for mainframes and servers to run complex AI models alongside demanding enterprise workloads, all while maintaining robust security and resilience.
Technically, the Spyre Accelerator is a sophisticated 5nm, 32-core System-on-a-Chip (SoC) featuring 25.6 billion transistors, packaged on a 75-watt PCIe card. Its design allows for impressive scalability, with the ability to cluster up to 48 cards in an IBM Z or LinuxONE system and 16 cards in an IBM Power system, enabling organizations to size deployments for multi-model serving, tenant isolation, and future growth without requiring platform redesigns. For Power systems, the Spyre Accelerator, combined with an on-chip accelerator (MMA), significantly accelerates data preparation and conversion, which are often bottlenecks in RAG (Retrieval Augmented Generation) and enterprise search applications.
“One of our key priorities has been advancing infrastructure to meet the demands of new and emerging AI workloads,” stated Barry Baker, COO, IBM Infrastructure & GM, IBM Systems. “With the Spyre Accelerator, we’re extending the capabilities of our systems to support multi-model AI – including generative and agentic AI. This innovation positions clients to scale their AI-enabled mission-critical workloads with uncompromising security, resilience, and efficiency, while unlocking the value of their enterprise data.”
The development of Spyre is the culmination of a decade of innovation from IBM Research’s AI Hardware Center, founded in 2019. Mukesh Khare, General Manager of IBM Semiconductors, highlighted the commercialization of the center’s first chip, designed to enhance performance for IBM’s mainframe and server clients. The accelerator evolved from a prototype chip, refined through rapid iteration and cluster deployments at IBM’s Yorktown Heights campus, and in collaboration with institutions like the University at Albany’s Center for Emerging Artificial Intelligence Systems.
Also Read:
- S&P Global and IBM Forge Alliance to Enhance Enterprise Operations with Agentic AI
- Lemony Unveils ‘AI in a Box’ for On-Premise Enterprise AI Solutions
General availability for the IBM Spyre Accelerator begins on October 28, 2025, for IBM z17 and LinuxONE 5 systems, with availability for Power11 servers following in early December 2025. This strategic move by IBM underscores its commitment to providing pragmatic paths for embedding generative and agentic AI directly into core enterprise systems, ensuring data locality, security, and operational efficiency.
