Details
- IBM will make the Spyre Accelerator available for its z17 mainframe on October 28, 2025, bringing powerful generative AI capabilities directly to enterprise mainframes.
- The Spyre Accelerator boasts 32 AI-optimized processing cores, is built using Samsung's 5nm technology with 25.6 billion transistors, and comes as a 75-watt PCIe card—supporting up to 48 cards per IBM Z or LinuxONE system.
- Paired with the new Telum II processor, featuring 8 cores running at 5.5GHz, 43 billion transistors, and 36MB of L2 cache per core, the system supports high-speed enterprise AI workloads and real-time fraud detection at a scale of 450 billion inference operations per day.
- Integration with IBM's watsonx Assistant for Z and other AI toolkits enables on-premises deployment of large language models in highly secure environments, a critical benefit for industries with strict compliance standards.
- IBM cites a 2024 Oxford Economics survey showing 61% of executives prioritize generative AI for mainframe modernization, with clients emphasizing the importance of keeping production data on-premises for security and regulatory needs.
Impact
IBM’s Spyre Accelerator and Telum II launch is a bold play to keep mainframes competitive in the AI era, directly targeting enterprises wary of cloud-based AI due to security and compliance. By enabling powerful AI workloads on-premises, IBM is meeting the demands of regulated industries and strengthening its foothold against cloud and accelerator rivals.