IBM's big iron to get Spyre AI accelerator upgrade this month (www.theregister.com)

IBM will begin shipping its Spyre AI accelerator this month, with general availability from Oct. 28 for z17 mainframes and LinuxONE 5 systems and early December for Power11 servers. Spyre is a PCIe accelerator card built around a custom chip with 32 dedicated AI cores (architecture akin to the Telum II on z17). Systems can be scaled to large accelerometer clusters—up to 48 cards on IBM Z/LinuxONE (1,536 accelerator cores) and up to 16 cards on Power systems (512 cores)—to run low‑latency inferencing alongside mission‑critical workloads. The significance is practical and strategic: IBM is embedding purpose‑built AI acceleration into enterprise big‑iron so customers can run generative AI, LLM inference and multi‑model predictive pipelines (e.g., fraud detection at transaction time) without offloading sensitive data. z/OS 3.2 adds native support for the accelerator and modern data access/NoSQL hooks to expose mainframe data to hybrid cloud and AI environments without heavy ETL. For organizations that require resilience, security and high throughput, Spyre enables on‑prem AI scaling with reduced latency and tighter integration into existing mainframe operations—extending IBM’s prior Telum efforts to meet growing AI inference demand in regulated, mission‑critical settings.