Power S1112 AI Inference for IBM i Shops
IBM's one-socket Power S1112 adds on-chip Matrix Math Acceleration for local AI inference. The fit depends on runtime, partition capacity, and the model, not an IBM i label alone.
Power S1112 is a one-socket Power 11 system IBM described as capable of local AI inference using on-chip Matrix Math Acceleration (MMA). It is a hardware option for a measured local inference workload. It is not evidence that IBM Bob, watsonx, or every IBM i application requires a new S1112.
What IBM actually announced
IBM's July 15, 2026 release introduced the compact S1112 and described faster local inferencing through Power 11 MMA. It gave July 24 as the expected general-availability date for the system. The same release discusses IBM Bob Premium Package for i and Power Autonomous Operations, but those are separate software offerings. A customer example mentions Linux partitions alongside IBM i, which is the operational distinction to keep in view.
Source: IBM Power S1112 announcement. Check IBM's current ordering and support documentation for actual availability and configuration details.
MMA is not a Spyre card
Two distinct acceleration options
| Technology | Infrastructure implication |
|---|---|
| Power 11 on-chip MMA | Processor-level matrix acceleration. The model, framework, supported runtime, cores, and memory still determine whether a local inference pilot fits. |
| IBM Spyre Accelerator for Power | PCIe-attached cards in supported Power 11 systems with an ENZ0 expansion drawer and a Red Hat inference environment. Confirm the exact machine and card configuration. |
Before choosing the S1112
Use the workload-placement guide before sizing. For Power 11 system specifications, continue to the Power 11 specialist. For AI coding tools such as IBM Bob, use the software AI site.