On-Chip AI Inference Silicon

IBM Telum II Processor

IBM's mainframe-class processor with a native, on-chip AI accelerator delivering 24 trillion operations per second -- four times the AI throughput of the original Telum chip.

Telum II is IBM's second-generation AI-integrated processor, built for the z17 mainframe. Where earlier AI accelerators bolted inference hardware onto a chip via memory-mapped I/O, Telum II's accelerator executes matrix multiplications and other AI primitives as native instructions -- cutting overhead and letting z17 systems handle millions of low-latency inference operations per transaction.

Telum II is not part of the Power processor family, and IBM has not stated that Power12 will use Telum II silicon directly. What IBM has confirmed is a related fact worth tracking closely: the Spyre Accelerator -- IBM's PCIe-based AI accelerator card built on the same generation of AI silicon research -- reached compatibility with Power11 systems in December 2025, and IBM distinguished engineer Bill Starke has said Spyre "will be a piece of the Power11 portfolio and beyond." That is the real, sourced link between Telum II-class AI silicon and where Power is headed.

This entry describes real, shipping IBM hardware (Telum II in z17, Spyre Accelerator support for Power11). No Power12-specific specification is implied or claimed.

Frequently Asked

How is Telum II related to Power 12?

Telum II is a mainframe processor (powering IBM's z17 system), not a Power chip, and IBM has not stated Power12 uses Telum II silicon. It matters here because it's the clearest public evidence of IBM's on-chip AI inference strategy, which has already extended into the Power line via the Spyre Accelerator's confirmed Power11 compatibility.