AI on IBM i: the Infrastructure Question
AI on IBM i can mean software that uses IBM i data or a model that runs on IBM Power hardware. Those are different decisions, often in different partitions.
'AI on IBM i' does not say where the AI actually runs. IBM Bob can assist with IBM i code while the underlying service remains a software decision. An agent can reach Db2 for i from another runtime. Local model inference is a hardware and partition decision. Ask where the workload executes before comparing Power systems.
The infrastructure choices
Existing system or external service
An IBM i data source can connect to AI software outside the production partition. Test the integration and data movement before sizing new hardware.
Power CPU inference
Power 11 on-chip Matrix Math Acceleration can help a suitable local model in a supported runtime. IBM's Power S1112 announcement makes this a concrete compact-system option.
Power Spyre acceleration
Spyre cards require a supported Power 11 system, ENZ0 drawer, and Red Hat inference environment. This is not the same as on-chip MMA or a native IBM i feature.
Storage and resilience
AI data pipelines and storage administration can affect latency, recoverability, and security. FlashSystem.ai detail belongs on the dedicated storage site.
Start with AI workload placement, then the infrastructure checklist. Software questions about watsonx, MCP connectivity, or IBM Bob belong on the software AI site. For FlashSystem.ai, use the FlashSystem specialist.
IBM's July 15, 2026 Power S1112 release is the source for on-chip local inference claims. Verify current system support and workload requirements with IBM before treating an announcement as a sizing recommendation.