IBM Announces Spyre Accelerator for Enterprise AI Workloads

IBM has announced the commercial availability of the Spyre Accelerator, an AI chip designed to handle generative and agentic AI workloads with low latency and high security. The accelerator will be available on October 28, 2025, for IBM z17 and LinuxONE 5 systems, and in early December for Power11 servers.

The Spyre Accelerator is built with 32 AI-optimized cores and 25.6 billion transistors, produced using 5-nanometer technology. Each unit is mounted on a 75-watt PCIe card, allowing up to 48 cards to be clustered in IBM Z or LinuxONE systems, or 16 cards in Power systems. This configuration enables enterprises to scale AI processing while keeping data on-premises.

In combination with the Telum II processor, the Spyre Accelerator supports high-volume transaction processing and real-time inference for tasks such as fraud detection and retail automation. The hardware is designed to execute multiple AI models simultaneously, maintaining low latency and throughput critical to enterprise operations.

The accelerator forms part of IBM’s broader infrastructure strategy to integrate AI directly into enterprise systems. It will also support on-premise deployment of tools such as watsonx Assistant for Z and Machine Learning for IBM z/OS, enabling organizations to modernize applications and maintain data security while adopting advanced AI capabilities.