Red Hat AI Enterprise explained

Red Hat AI Enterprise is Red Hat's all-in-one AI platform, announced on 24 February 2026. Red Hat describes it as "an integrated AI platform for deploying and managing AI models, agents and applications across the hybrid cloud", and its AI chief called it a "unified metal to agent platform". The phrase is the point: one supported stack from the operating system on the server to the agent a user talks to.

This article explains what is inside it, what it does with data, how it pairs with NVIDIA, and what it means for a bank, a ministry or a force planning AI on its own premises. It is not a product called "Red Hat AI Data Platform"; no such product exists, and the confusion is common enough that we address it below.

What it brings together

Red Hat AI Enterprise sits on Red Hat Enterprise Linux and Red Hat OpenShift, and joins the products Red Hat already sold for AI:

Product	Role
Red Hat AI Inference Server	Serves models fast and efficiently. Built on vLLM, the open-source inference engine Red Hat now leads the development of.
Red Hat OpenShift AI	The platform for the AI lifecycle: data science pipelines, training and tuning, model serving, monitoring, and since version 3.4 a catalog for the Model Context Protocol tools that agents use.
Red Hat Enterprise Linux AI	A bootable image of RHEL with the inference stack and NVIDIA, AMD or Intel accelerator support, for single-server deployments.

The press release leaves one thing deliberately vague: which pieces are bundled inside Red Hat AI Enterprise and which are sold alongside it. In practice OpenShift is at the centre, and the rest attaches to it.

The same day: Red Hat AI 3.3

Red Hat AI 3.3 shipped with the launch. Its headline features:

• A "unified data-to-model lifecycle" backed by a new Red Hat AI Python Index: hardened, supported builds of Docling (document conversion for RAG), SDG Hub (synthetic data generation) and Training Hub (model tuning), so pipelines can be repeatable and security-reviewed.
• Technology previews of Models-as-a-Service, generative AI on Intel CPUs, and NVIDIA NeMo Guardrails.
• Hardware certification for NVIDIA Blackwell Ultra and support for AMD MI325X accelerators.

At Red Hat Summit in May 2026, Red Hat AI 3.4 followed with a focus on agentic workflows: the MCP catalog in OpenShift AI, Red Hat AI Inference running on any managed Kubernetes service with launch blueprints for CoreWeave and Azure, and a partnership with Boomi for deploying agents.

Red Hat AI Factory with NVIDIA

Red Hat and NVIDIA co-engineered a combined offering: Red Hat AI Enterprise plus NVIDIA AI Enterprise, sold as the Red Hat AI Factory with NVIDIA. NVIDIA supplies the GPU drivers, NIM inference microservices, NeMo tools and the AI Blueprints; Red Hat supplies the operating system, the Kubernetes platform and the lifecycle tooling. This is the stack our lab runs, and the one behind our sovereign AI proposals: OpenShift AI for the platform, NVIDIA AI Enterprise for the models and the GPU software, on servers from Dell or HPE.

What it does, and does not, do with data

Red Hat AI Enterprise is an AI platform, not a data platform. Its data features sit at the edges of the model lifecycle:

• Docling converts PDFs, scans and Office files into clean text for retrieval, and Red Hat documents running it at scale with Ray on OpenShift AI.
• SDG Hub generates synthetic training data; Training Hub tunes models on it.
• Kafka streaming comes from a separate product, Red Hat Streams for Apache Kafka; storage from OpenShift Data Foundation or partner storage such as NetApp and Dell; databases from the operators of Crunchy Data, Oracle or others.

The lakehouse, the catalogue, the vector index and the pipelines that keep them fresh are a separate design decision. That is why we treat the data platform as its own service: Red Hat AI Enterprise runs the models; something else has to run the data. Red Hat's larger role in data is underneath: IBM watsonx.data and Cloudera both run on OpenShift on premises.

Where it fits for institutions here

Three properties matter most for a bank, a ministry or the armed forces:

1. It runs on premises and air-gapped. Every component installs from a mirrored registry with no call home, which is the condition for sovereign deployments and the way our lab is built.
2. It is one support contract from the OS to the model. When a GPU driver, a Kubernetes upgrade and an inference server disagree, one vendor owns the answer.
3. The hard parts are open source. vLLM, Docling, KServe and Kubeflow are open projects, so skills and workloads move; the subscription buys the integration, the hardening and the support.

What it does not remove is the data work. The institutions whose AI projects stall usually have a model problem that turns out to be a data problem. Red Hat AI Enterprise makes the model side routine; the data side is where the engineering effort goes.

Red Hat's partner tiers, for reference

Red Hat's partner programme has three tiers, Ready, Advanced and Premier, assigned on engagement points, certifications and trained sellers. The tier is the vendor's to award and belongs in exact words next to its logo. It is not a general description of a partnership.