IBM Open-Source AI Foundation for Power
Supplies the ready-made, containerized AI services an IBM i shop deploys on a Spyre-equipped Power server, ordered from IBM as program number 5639-SAI. IBM lists it next to Red Hat AI Inference Server as a required part of a typical Spyre inference setup, so it is a line on the order rather than an optional download.
What does it actually give you?
Think of buying a kitchen. Red Hat AI Inference Server is the stove, and the Spyre Accelerator card is the gas line. This product is the crate of finished appliances that plug into both.
IBM calls each one an AI Service. The January 2026 Redbook describes a service as a self-contained unit that bundles the model, the runtime settings and an interface other programs call. You do not pick a model, download weights, guess at settings and wire up a server yourself. You deploy a container that already does one job.
- It is part of the IBM Spyre for Power portfolio. The Redbook names four products in that portfolio: this service catalog, Red Hat AI Inference Server on IBM Power as the inferencing platform, IBM Spyre Enablement Stack for Power as the accelerated infrastructure layer, and IBM Spyre Stack Updates for Power for ongoing driver, firmware and model updates.
- It also carries the Spyre acceleration utilities. The Redbook tells you to install those utilities from this product, and to confirm the Spyre device driver and firmware are in place before you expect any acceleration.
- It includes a Python side. IBM ships Python wheels for object detection and image classification, plus document handling such as PDF to text and optical character recognition, and the current documentation set pairs the AI Services with a Python Ecosystem for IBM Power section.
- IBM supports the code. The Redbook says the Spyre on Power customer support plan covers the Python packages released as part of the IBM Open-Source AI Foundation for Power repository, which matters because open source normally means you are on your own.
Do you need it to run a Spyre card?
By IBM's own description of a working setup, yes. The Redbook spells out what a typical Spyre inference environment looks like: a Power 11 host with eight or more Spyre cards in an ENZ0 PCIe4 expansion drawer, running Red Hat Enterprise Linux 9.6 or later or Red Hat OpenShift, and including two software items. One is Red Hat AI Inference Server. The other is IBM Open-Source AI Foundation for Power.
That is worth knowing before you price a card. The accelerator is the part people quote. The software beside it is a separate order, with its own program number and its own term.
The Redbook describes a reference configuration, not a licensing contract. IBM sells Red Hat AI Inference Server, the enablement stack and this product under separate program numbers, and the support and subscription terms have already changed twice since launch. Confirm what your own order has to include with IBM or your reseller before you budget.
Which jobs does it cover?
The Redbook lists the work the services handle at release, built on IBM Granite and other supported open models. The current documentation set then names the finished services you install.
The jobs IBM lists for AI Services at release, and what each one is for
| Job | What it does | Where it is useful |
|---|---|---|
| Text generation | Writes, rewrites and summarizes text, and answers questions in conversation, using Granite instruct models. | A help desk reply, or a summary of a long order note. |
| Embedding generation | Turns a document into a list of numbers, so a computer can find things that mean the same thing instead of things spelled the same. | Searching a knowledge base by meaning. |
| Retrieval-augmented generation | Looks up your own documents first, then writes an answer from what it found. Shortened to RAG. | Answering from your own policies instead of from the open internet. |
| Re-ranking | Takes search results that already came back and reorders them by how well they really fit. | Putting the right answer at the top of a long result list. |
| Entity extraction | Pulls structured fields out of messy text, such as a name, an amount or a date. | Reading an invoice or a scanned form into a database. |
Jobs from section 4.2.3 of the IBM Redbook Implementing AI on Power 11, SG24-8592-00, January 2026. The current IBM documentation set, SSXZBY_2026.06.0, names the installable services as Digitize Documents, Question and Answer, Find Similar Items and Summarize.
When did IBM ship it?
Three IBM announcements that changed what you can buy and where it runs
- 12 December 2025Version 1.0.0 becomes generally available as program number 5639-SAI under announcement AD25-1365, licensed under the International Program License Agreement. Power 11 only at this point.
- April 2026Announcement AD26-0442 makes it available on Power 10 technology-based servers. An existing Power 10 shop no longer has to replace the server first.
- June 2026Announcement AD26-0481 reinstates the 5-year program options for Red Hat AI Inference and for IBM Open-Source AI Foundation for Power, so a longer term is back on the price list.
Dates and announcement numbers from the IBM Power announcements index and the IBM support lifecycle page for IBM Open-Source AI Foundation for Power 1.0.0.
How do you deploy it?
The services arrive as containers, so the real question is what runs the containers. IBM documents three ways, and they line up with how serious the workload is.
Three ways IBM documents for running an AI Service
- 1. Podman on its own, for trying thingsStart the container by hand on a Red Hat Enterprise Linux partition. IBM points this at development, testing and light use. The documentation gives Spyre chip alignment inside the partition its own page, because that is the step people get wrong.
- 2. Podman Quadlets, for production on one partitionA Quadlet hands the container to systemd, the part of Linux that starts and restarts services. IBM recommends this when you want the service to come back after a reboot without anyone typing a command.
- 3. Red Hat OpenShift, for more than oneIBM ships a small command line installer, version 0.3.2 in the June 2026 documentation, that sets up what OpenShift needs: IBM Spyre support, Red Hat OpenShift AI, hardware feature detection, certificate handling, service mesh and scheduling. Pick this when several models, teams or servers are involved.
Deployment options from section 4.2.4 of the IBM Redbook and from the OpenShift setup topic in IBM documentation set SSXZBY_2026.06.0. That documentation also covers IBM Power Virtual Server and IBM watsonx as targets.
Related Terms
IBM Spyre Enablement Stack for Power
The drivers, runtime and models IBM requires before a Spyre Accelerator card runs anything on a Power 11 server. It arrives from Red Hat inside Red Hat AI Inference Server, so the card sits idle until the Linux partition that drives it has the stack installed and entitled.
Red Hat AI Inference Server
Red Hat's model serving product, built on the open source vLLM engine. IBM lists it as one of two supported ways to drive a Spyre Accelerator on IBM Power.
Spyre Accelerator
IBM's PCIe AI inference card. On IBM Power 11 it carries 32 cores, 128 GB of memory, and more than 300 TOPS inside a 75 watt envelope.
IBM Granite
IBM's own family of open language models, published under the Apache 2.0 licence so they can be downloaded, tuned, and run on hardware you own.
Red Hat OpenShift AI
Red Hat's larger AI platform, and the second option IBM accepts for driving a Spyre Accelerator on IBM Power.